This book is in the

ADDISON-WESLEY SERIES IN MATHEMATICS

Consulting Editor: LYNN H. LOOMIS

Introduction to

ComIllutative Algebra

M. F. ATIYAH, FRS I UNIVERSITY OF OXFORD I. G. MACDONALD

ADDISON-WESLEY PUBLISHING COMPANY

Reading, Massachusetts Menlo Park, California . London . Don Mills, Ontario

Copyright © 1969 by Addison-Wesley Publishing Company, Inc.

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without prior written permission of the publisher. Printed in Great Britain. Library of Congress Catalog Card No. 72-79530.

Contents

Introduction .

Notation and Terminology

Chapter 1 Rings and Ideals.

Rings and ring homomorphisms Ideals. Quotient rings . . Zero-divisors. Nilpotent elements. Units. Prime ideals and maximal ideals NiIradical and Jacobson radical Operations on ideals Extension and contraction. Exercises

Chapter 2 Modules.

Modules and module homomorphisms Submodules and quotient modules Operations on submodules. Direct sum and product Finitely generated modules Exact sequences. Tensor product of modules Restriction and extension of scalars Exactness properties of the tensor product Algebras. Tensor product of algebras Exercises.

Chapter 3 Rings and Modules of Fractions

Local properties. Extended and contracted ideals in rings of fractions Exercises.

Chapter 4 Primary Decomposition

Exercises.

v

vii

ix

1 2 2 3 5 6 9

10

17

17 18 19 20 21 22 24 27 28 29 30 31

36

40 41 43

50

55

vi CONTENTS

Chapter 5 Integral Dependence and Valuations 59

Integral dependence 59 The going-up theorem 61 Integrally closed integral domains. The going-down theorem. 62 Valuation rings 65 Exercises. 67

Chapter 6 Chain Conditions 74

Exercises. 78

Chapter 7 Noetherian Rings 80

Primary decomposition in Noetherian rings 82 Exercises. 84

Chapter 8 Artin Rings 89

Exercises. 91

Chapter 9 Discrete Valuation Rings and Dedekind Domains

Discrete valuation rings Dedekind domains Fractional ideals Exercises.

Chapter 10 Completions.

Topologies and completions Filtrations Graded rings and modules. The associated graded ring Exercises.

Chapter 11 Dimension Theory

Hilbert functions Dimension theory of Noetherian local rings Regular local rings . Transcendental dimension. Exercises.

Index.

93

94 95 96 99

100

101 105 106 III 113

116

116 119 123 124 125

127

Introduction

Commutative algebra is essentially the study of commutative rings. Roughly speaking, it has developed from two sources: (I) algebraic geometry and (2) algebraic number theory. In (1) the prototype of the rings studied is the ring k[Xb ... , xn] of polynomials in several "ariables over a field k; in (2) it is the ring Z of rational integers. Of these two the algebro-geometric case is the more far-reaching and, in its modern development by Grothendieck, it embraces much of algebraic number theory. Commutative algebra is now one of the foundation stones of this new algebraic geometry. It provides the complete local tools for the subject in much the same way as differential analysis provides the tools for differential geometry.

This book grew out of a course of lectures given to third year under­graduates at Oxford University and it has the modest aim of providing a rapid introduction to the subject. It is designed to be read by students who have had a first elementary course in general algebra. On the other hand, it is not intended as a substitute for the more voluminous tracts on commutative algebra such as Zariski-Samuel [4] or Bourbaki [1]. We have concentrated on certain central topics, and large areas, such as field theory, are not touched. In content we cover rather more ground than Northcott [3] and our treatment is substantially different in that, following the modern trend, we put more emphasis on modules and localization.

The central notion in commutative algebra is that of a prime ideal. This provides a common generalization of the primes of arithmetic and the points of geometry. The geometric notion of concentrating attention "near a point" has as its algebraic analogue the important process of localizing a ring at a prime ideal. r t is not surprising, therefore, that results about localization can usefully be thought of in geometric terms. This is done methodically in Grothendieck's theory of schemes and, partly as an introduction to Grothendieck's work [2], and partly because of the geometric insight it provides, we have added schematic versions of many results in the form of exercises and remarks.

The lecture-note origin of this book accounts for the rather terse style, with little general padding, and for the condensed account of many proofs. We have resisted the temptation to expand it in the hope that the brevity of our presenta­tion will make clearer the mathematical structure of what is by now an elegant

vii

viii INTRODUCTION

and attractive theory. Our philosophy has been to build up to the main theorems in a succession of simple steps and to omit routine verifications.

Anyone writing now on commutative algebra faces a dilemma in connection with homological algebra, which plays such an important part in modern developments. A proper treatment of homological algebra is impossible within the confines of a small book: on the other hand, it is hardly sensible to ignore it completely. The compromise we have adopted is to use elementary homological methods--exact sequences, diagrams, etc.-but to stop short of any results requiring a deep study of homology. In this way we hope to prepare the ground for a systematic course on homological algebra which the reader should under­take if he wishes to pursue algebraic geometry in any depth.

We have provided a substantial number of exercises at the end of each chapter. Some of them are easy and some of them are hard. Usually we have provided hints, and sometimes complete solutions, to the hard ones. We are indebted to Mr. R. Y. Sharp, who worked through them all and saved us from error more than once.

We have made no attempt to describe the contributions of the many mathematicians who have helped to develop the theory as expounded in this book. We would, however, like to put on record our indebtedness to J.-P. Serre and J. Tate from whom we learnt the subject, and whose influence was the determining factor in our choice of material and mode of presentation.

REFERENCES

1. N. BOURBAKI, Algebre Commutative, Hermann, Paris (1961-65).

2. A. GROTHENDIECK and J. DIEUDONNE, Elements de Geometrie Algebrique, Publications Mathematiques de /' I.H.E.S., Nos. 4, 8, 11, .. , Paris (1960-- ).

3. D. G. NORTHcorr, Ideal Theory, Cambridge University Press (1953).

4. O. ZARISKI and P. SAMUEL, Commutative Algebra I, II, Van Nostrand, Princeton (1958, 1960).

Notation and Terminology

Rings and modules are denoted by capital italic letters, elements of them by small italic letters. A field is often denoted by k. Ideals are denoted by small German characters. Z, Q, R, C denote respectively the ring of rational integers, the field of rational numbers, the field of real numbers and the field of complex numbers.

Mappings are consistently written on the left, thus the image of an element x under a mappingfis written f(x) and not (x)! The composition of mappings f: X --+ Y, g: Y -+ Z is therefore go J, notf 0 g.

A mappingf: X -+ Yis injective iff(xl) = f(X2) implies Xl = X2; surjective if f(X) = Y; bijective if both injective and surjective.

The end of a proof (or absence of proof) is marked thus •.

Inclusion of sets is denoted by the sign s;;. We reserve the sign c: for strict inclusion. Thus A c: B meanS that A is contained in B and is not equal to B.

ix

1

Rings and Ideals

We shall begin by reviewing rapidly the definition and elementary properties of rings. This will indicate how much we are going to assume of the reader and it will also serve to fix notation and conventions. After this review we pass on to a discussion of prime and maximal ideals. The remainder of the chapter is devoted to explaining the various elementary operations which can be performed on ideals. The Grothendieck language of schemes is dealt with in the exercises at the end.

RINGS AND RING HOMOMORPHISMS

A ring A is a set with two binary operations-(addition and multiplication) such that

1) A is an abelian group with respect to addition (so that A has a zero element, denoted by 0, and every x E A has an (additive) inverse, -x).

2) Multiplication is associative (xy)z = x(yz)) and distributive over addition (x(y + z) = xy + xz, (y + z)x = yx + zx).

We shall consider only rings which are commutative:

3) xy = yx for all x, YEA,

and have an identity element (denoted by 1):

4) 31 E A such that xl = Ix = x for all x E A. The identity element is then unique.

Throughout this book the word "ring" shall mean a commutative ring with an identity element, that is, a ring satisfying axioms (1) to (4) above.

Remark. We do not exclude the possibility in (4) that 1 might be equal to O. If so, then for any x E A we have

x = xl = xO = 0

ar{d so A has only one element, o. In this caSe A is the zero ring, denoted by 0 (by abuse of notation).

2 RINGS AND IDEALS

A ring homomorphism is a mappingf of a ring A into a ring B such that

i) f(x + y) = f(x) + f(y) (so thatfis a homomorphism of abelian groups, and therefore alsof(x - y) = f(x) - f(y),f( -x) = -f(x),f(O) = 0),

ii) f(xy) = f(x)f(y),

iii) f(1) = 1.

In other words,frespects addition, multiplication and the identity element.

A subset S of a ring A is a subring of A if S is closed under addition and multiplication and contains the identity element of A. The identity mapping of S into A is then a ring homomorphism.

Iff: A -+ B, g: B -+ C are ring homomorphisms then so is their composition gof: A -+ C.

IDEALS. QUOTIENT RINGS

An ideal a of a ring A is a subset of A which is an additive subgroup and is such that Aa s; a (i.e., x E A and YEa imply xy E a). The quotient group A/a inherits a uniquely defined multiplication from A which makes it into a ring, called the quotient ring (or residue-class ring) A/a. The elements of A/a are the co sets of a in A, and the mapping cp: A -+ A/a which maps each x E A to its coset x + a is a surjective ring homomorphism.

We shall frequently use the following fact:

Proposition 1.1. There is a one-to-one order-preserving, correspondence between the ideals b of A which contain a, and the ideals b of A/a, given by b = cp-l(b). •

Iff: A -+ B is any ring homomorphism, the kernel off( = f-l(O) is an ideal a of A, and the image of f( = f(A) is a subring C of B; and f induces a ring isomorphism A/a ~ C.

We shall sometimes use the notation x == y (mod a); this means that x - yEa.

ZERO-DIVISORS. NILPOTENT ELEMENTS. UNITS

A zero-divisor in a ring A is an element x which "divides 0", i.e., for which there exists y '# 0 in A such that xy = O. A ring with no zero-divisors '# 0 (and in which I '# 0) is called an integral domain. For example, Z and k[xlo ... , x,,] (k a field, XI indeterminates) are integral domains.

An element x E A is nilpotent if x" = 0 for some n > O. A nilpotent element is a zero-divisor (unless A = 0), but not conversely (in general).

A unit in A is an element x which "divides I ", i.e., an element x such that xy = 1 for some YEA. The element y is then uniquely determined by x, and is written x- l

• The units in A form a (multiplicative) abelian group.

PRIME IDEALS AND MAXIMAL IDEALS 3

The multiples ax of an element x E A form a principal ideal, denoted by (x) or Ax. x is a unit <::> (x) = A = (1). The zero ideal (0) is usually denoted by O.

A field is a ring A in which 1 '# 0 and every non-zero element is a unit. Every field is an integral domain (but not conversely: Z is not a field).

Proposition 1.2. Let A be a ring '# o. Then the following are equivalent: i) A is a field;

ii) the only ideals in A are 0 and (1);

iii) every homomorphism of A into a non-zero ring B is injective.

Proof i) => ii). Let a '# 0 be an ideal in A. Then a contains a non-zero element x; x is a unit, hence a ;;2 (x) = (1), hence a = (1).

ii) => iii). Let 4>: A --+ B. be a ring homomorphism. Then Ker (4)) is an ideal '# (1) in A, hence Ker (4)) = 0, hence 4> is injective.

iii) => i). Let x be an element of A which is not a unit. Then (x) '# (1), hence B = A/(x) is not the zero ring. Let 4>: A -?- B be the natural homo­morphism of A onto B, with kernel (x). By hypothesis, 4> is injective, hence (x) = 0, hence x = O. •

PRIME IDEALS AND MAXIMAL IDEALS

An ideal V in A is prime if V '# (1) and if xy E V => X E V or y E V. An ideal m in A is maximal if m '# (1) and if there is no ideal a such that

mea c: (1) (strict inclusions). Equivalently:

V is prime <::> A/V is an integral domain; m is maximal <::> A/m is a field (by (1.1) and (1.2».

Hence a maximal ideal is prime (but not conversely, in general). The zero ideal is prime <::> A is an integral domain.

If f: A --+ B is a ring homomorphism and q is a prime ideal in B, then f-l( q) is a prime ideal in A, for AIf-l( q) is isomorphic to a subring of B/q and hence has no zero-divisor '# O. But if n is a maximal ideal of B it is not neces­sarily trye thatf-l(n) is maximal in A; all we can say for sure is that it is prime. (Example: A = Z, B = Q, n = 0.)

Prime ideals are fundamental to the whole of commutative algebra. The following theorem and its corollaries ensure that there is always a sufficient supply of them.

Theorem 1.3. Every ring A '# 0 has at least one maximal ideal. (Remember that "ring" means commutative ring with 1.)

Proof This is a standard application of Zorn's lemma. * Let ~ be the set of all ideals '# (1) in A. Order ~ by inclusion. ~ is not empty, since 0 E~. To apply

• Let S be a non-empty partially ordered set (i.e., we are given a relation x ~ yon S which is reflexive and transitive and such that x ~ y and y ~ x together imply

4 RINGS AND IDEALS

Zorn's lemma we must show that every chain in ~ has an upper bound in ~; let then (aa) be a chain of ideals in~, so that for each pair of indices IX, f3 we have either aa s; ap or ap S; aa' Let a = Ua aa. Then a is an ideal (verify this) and 1 ¢ a because I ¢ aa for all IX. Hence a E~, and a is an upper bound of the chain. Hence by Zorn's lemma ~ has a maximal element. _

Corollary 1.4. If a :f. (1) is an ideal of A, there exists a maximal ideal of A containing a.

Proof Apply (1.3) to A/a, bearing in mind (1.1). Alternatively, modify the proof of (I.3). _

Corollary 1.5. Every non-unit of A is contained in a maximal ideal. _

Remarks. 1) If A is Noetherian (Chapter 7) we can avoid the USe of Zorn's lemma: the set of all ideals :f. (1) has a maximal element.

2) There exist rings with exactly one maximal ideal, for example fields. A ring A with exactly one maximal ideal m is called a local ring. The field k = A/m is called the residue field of A.

Proposition 1.6. i) Let A be a ring and m :f. (1) an ideal of A such that every x E A - m is a unit in A. Then A is a local ring and m its maximal ideal.

ii) Let A be a ring and m a maximal ideal of A, such that every element of 1 + m (i.e., every 1 + x, where x E m) is a unit in A. Then A is a local ring.

Proof i) Every ideal :f. (1) consists of non-units, hence is contained in m. Hence m is the only maximal ideal of A.

ii) Let x E A - m. Since m is maximal, the ideal generated by x and m is (I), hence there existy E A and t E m such that xy + t = 1; hence xy = 1 - t belongs to 1 + m and therefore is a unit. Now use i). _

A ring with only a finite number of maximal ideals is caIled semi-local.

Examples. 1) A = k[Xb ' . " xn ], k a field. Let f E A be an irreducible poly­nomial. By unique factorization, the ideal (f) is prime.

2) A = Z. Every ideal in Z is of the form (m) for some m ~ O. The ideal (m) is prime -¢> m = 0 or a prime number. All the ideals (p), where p is a prime number, are maximal: Z/(P) is the field of p elements.

The same holds in Example 1) for n = 1, but not for n > 1. The ideal m of all polynomials in A = k[Xl, ' , " x n ] with zero constant term is maximal (since

x = y). A subset T of S is a chain if either x .s;; y or y .s;; x for every pair of elements x, yin T. Then Zorn's lemma may be stated as follows: if every chain T of S has an upper bound in S (i.e" if there exists XES such that t .s;; x for all t E T) then S has at least one maximal element.

For a proof of the equivalence of Zorn's lemma with the axiom of choice, the well-ordering principle, etc., See for example P •. R. Halmos, Na(ve Set Theory, Van Nostrand (1960).

NILRADICAL AND JACOBSON RADICAL 5

it is the kernel of the homomorphism A --+ k which maps f E A to f(O»). But if n > I, m is not a principal ideal: in fact it requires at least n generators.

3) A principal ideal domain is an integral domain in which every ideal is principal. In such a ring every non-zero prime ideal is maximal. For if (x) '# 0 is a prime ideal and (y) ::::> (x), we have x E (y), say x = yz, so that yz E (x) and y ¢ (x), hence Z E (x): say Z = tx. Then x = yz = ytx, so that yt = I and therefore (y) = (1).

NILRADICAL AND JACOBSON RADICAL

Proposition 1.7. The set 91 of all nilpotent elements in a ring A is an ideal, and A/91 has no nilpotent element '# O.

Proof If x E 91, clearly ax E 91 for all a EA. Let x, y E 91: say xm = 0, yn = O. By the binomial theorem (which is valid in any commutative ring), (x + y)m+n-l is a sum of integer multiples of products xry', where r + s = m + n - I; we cannot have both r < m and s < n, hence each of these products vanishes and therefore (x + y)m+n-l = O. Hence x + y E 91 and therefore 91 is an ideal.

Let ,i E A/91 be represented by x E A. Then,in is represented by x n, so that ,in = 0 => xn E 91 => (xn)k = 0 for some k > 0 => X E 91 => .'\: = O. •

The ideal 91 is called the nilradical of A. The following proposition gives an alternative definition of 91:

Proposition 1.8. The nilradical of A is the intersection of all the prime ideals ofA.

Proof Let 91' denote the intersection of all the prime ideals of A. Iff E A is nilpotent and if p is a prime ideal, then r = 0 E P for some n > 0, hence fE p (because p is prime). HencefE91'.

Conversely, suppose that f is not nilpotent. Let ~ be the set of ideals a with the property

n > 0 => fn ¢ u.

Then ~ is not empty because 0 E~. As in (1.3) Zorn's lemma can be applied to the set ~, ordered by inclusion, and therefore ~ has a maximal element. Let p be a maximal element of~. We shall show that p is a prime ideal. Let x, y ¢ p. Then the ideals p + (x), p + (y) strictly contain p and therefore do not belong to ~; hence

fm E P + (x), rEp + (y)

for some m, n. It follows thatfm+n E p + (xy), hence the ideal p + (xy) is not in ~ and therefore xy ¢ p. Hence we have a prime ideal p such thatf ¢ p, so that f¢ 91'. •

The Jacobson radical m of A is defined to be the intersection of all the maxi­mal ideals of A. It can be characterized as follows:

6 RINGS AND IDEALS

Proposition 1.9. x E m <::> 1 - xy is a unit in A for all YEA.

Proof :::;.: Suppose 1 - xy is not a unit. By (1.5) it belongs to some maximal ideal m; but x EmS;; m, hence xy E m and therefore 1 Em, which is absurd.

¢:: Suppose x ¢ m for some maximal ideal m. Then m and x generate the unit ideal (1), so that we have u + xy = 1 for some u E m and some YEA. Hence 1 - xy E m and is therefore not a unit. _

OPERATIONS ON IDEALS

If a, 0 are ideals in a ring A, their sum a + 0 is the set of all x + y where x E a and YEO. It is the smallest ideal containing a and O. More generally, we may define the sum 2:lel ~ of any family (possibly infinite) of ideals a, of A; its ele­ments are all sums 2: XI> where XI E al for all i E I and almost all of the XI (i.e., all but a finite set) are zero. It is the smallest ideal of A which contains all the ideals a l .

The intersection of any family (~)Iel of ideals is an ideal. Thus the ideals of A form a complete lattice with respect to inclusion.

The product of two ideals a, 0 in A is the ideal ao generated by all products xy, where x E a and YEO. It is the set of all finite sums 2: XIYI where each XI E a and each YI EO. Similarly we define the product of any finite family of ideals. In particular the powers an (n > 0) of an ideal a are' defined; conventionally, aD = (1). Thus an (n > 0) is the ideal generated by all products XiX:!· •• Xn

in which each factor XI belongs to a.

Examples. 1) If A = Z, a = (m), 0 = (n) then a + 0 is the ideal generated by the h.c.f. of m and n; a (") 0 is the ideal generated by their l.c.m.; and ao = (mn). Thus (in this case) ao = a (") 0 <::> m, n are coprime.

2) A = k[Xl, ••. , xn], a = (Xl, . .. , Xn) = ideal generated by Xl>"" Xn•

Then am is the set of all polynomials with no terms of degree < m.

The three operations so far defined (sum, intersection, product) are all commutative and associative. Also there is the distributive law

a(o + c) = ao + ac.

In the ring Z, nand + are distributive over each other. This is not the case in general, and the best we have in. this direction is the modular law

a n (0 + c) = a (") 0 + a (") c if a ;:;2 0 or a ;:;2 c.

Again, in Z, we have (a + o)(a no) = ao; but in general we have only (a + o)(a (") 0) s;; ao (since (a + o)(a (") 0) = a(a n 0) + o(a (") 0) S;; ao). Clearly ao S;; an 0, hence

a (") 0 = ao provided a + 0 = (l).

OPERA nONS ON IDEALS 7

Two ideals a, 0 are said to be coprime (or comaximal) if a + b = (I). Thus for coprime ideals we have a (") b = ao. Clearly two ideals a, b are coprime if and only if there exist x E a and YEO such that x + Y = 1.

Let Ai> ... , An be rings. Their direct product

n

A =nAj i= 1

is the set of all sequences x = (Xl>' .. , xn) with Xj E Aj (1 .;;:; i .;;:; n) and com­ponentwise addition and m\lltiplication. A is a commutative ring with identity element (1,1, ... ,1). We have projections Pj: A -+ Ai defined by pj(x) = Xj; they are ring homomorphisms.

Let A be a ring and ab ... , an ideals of A. Define a homomorphism

n

cp: A -+ n (A/a j )

j= 1

by the rule cp(x) = (x + a1 , ... , x + an).

Proposition 1.10. i) If a;, a j are coprime whenever i "# j, then TIa j = n a;.

ii) cp is surjective <=> aj, aj are coprime whenever i "# j.

iii) cp is injective <=> n a j = (0).

Proof i) by induction on n. The case n = 2 is dealt with above. Suppose n > 2 and the result true for al> ... , an - l> and let 0 = n~;l a j = n~.;1 a j • Since a j + an = (1) (1 .;;:; i .;;:; n - 1) we have equations Xj + Yj = 1 (Xj E aj, Yj E an) and therefore

n-1 n-1 n Xj = n (1 - Yj) == 1 (mod an). j= 1 j = 1

Hence an + 0 = (I) and so

ii) =>: Let us show for example that a1, a:l are coprime. There exists x E A such that cp(x) = (1,0, ... ,0); hence x == 1 (mod a1) and x == 0 (mod a2), so that

1 = (1 - x) + X E a1 + a:l'

<=: It is enough to show, for example, that there is an element x E A such that q,(x) = (1,0, ... , 0). Since a 1 + a, = (1) (i > I) we have equations u, + v, = 1 (u, E al> v, E a,). Take x = f1~=:1 Vj, then x = TI(I - Uj) == I (mod a1), and x == ° (mod a,), i > 1. Hence cp(x) = (1,0, ... ,0) as required.

iii) Clear, since n a j is the kernel of cpo •

The union a u 0 of ideals is not in general an ideal.

8 RINGS AND IDEALS

Proposition 1.11. i) Let .\.1b ... , Pn be prime ideals and let a be an ideal contained in Uf= 1 Vj· Then a S; VI for some i.

ii) Let ab ... , an be ideals and let V be a prime ideal containing nf~ 1 ai ·

Then V :? aj for some i. If P = n at. then V = ai for some i.

Proof i) is proved by induction On n in the form n

a $ .\.1i (l ~ i ~ n) => a $ U lJi' j = 1

It is certainly true for n = 1. If n > I and the result is true for n - 1, then for each i there exists Xi E a such that Xi ¢: .\.1f whenever j =f. i. If for some i we have' Xi ¢ VI. we are through. If not, then Xi E Pi for all i. Consider the element

n

y = L: XIX2 " 'Xi-l X i+l X j+2" ·Xn ;

i= 1

we have YEa and) ¢:.\.1i (1 ~ i ~ n). Hence a $ Uf=l.\.1i'

ii) Suppose 1> ~r Qi for all i. Then there exist Xl E at> Xi ¢: V (l ~ i ~ n), and therefore TIxj E TIai S; n aj; but TIXi ¢: V (since V is prime). Hence V ;j2 n ai •

Finally, if V = n llj, then t> S; a; and hence V = a i for some i. •

If a, 0 are ideals in a rIng A, their Ideal quotient is

(a:o) = {xEA:xo S; a}

which is an ideal. In particular, (0: 0) is called the annihilator of 0 and is also denoted by Ann (0): it is the set of all X E A such that xo = O. In this notation the set of all zero-divisors in A is

D = U Ann (x). x,.o

If 0 is a principal ideal (x), we shall write (a : x) in place of (a : (x».

Example. If A = Z, a = (m), 0 = (n), where say m = Dp pltp, n = Dp pVP, then (a:o) = (q)whereq = DppYp and

yp = max (J.Lp - lip, 0) = J.Lp - min (J.Lp, lip).

Hence q = m/(m, n), where (m, n) is the h.c.f. of m and n.

Exercise 1.11. i) a S; (a: 0) ii) (a:o)o S; a

iii) (a: 0) : c) = (a: 0 c) = (a: c): 0 ) iv) (n, aj:o) = n (aj:o) v) (a:ZI Oi) = n (a:o j ).

If a is any ideal of A, the radical of a is

rea) = {x E A :xn E a for some n > O}.

If</>: A ~ A/a is the standard homomorphism, then rea) = q,-1(IJlA1a) and hence rea) is an ideal by (1.7).

Exercise 1.13. i) rea) ~ a

ii) r(r(a)) = rea) iii) r(ab) = rea n b) = rea) n reb) iv) rea) = (1) <0> a = {I) v) rea + b) = r(r(a) + reb»~

EXTENSION AND CONTRACTION 9

vi) if p is prime, r(pn) = pior all n > O.

Proposition 1.14. The radical of an ideal a is the intersection of the prime ideals which contain a.

Proof Apply (1.8) to A/a. _

More generally, we may define the radical r(E) of any subset E of A in the same way. It is not an ideal in general. We have r(U" E,,) = U r(E,,), for any family of subsets E" of A.

Proposition 1.15. D = set of zero-divisors of A = UX,.o r(Ann (x».

Proof D = reD) = r(UX,.o Ann (x» = Ux,.o r(Ann (x». _

Example. If A = Z, a = (m), let Pi (1 ~ i ~ r) be the distinct prime divisors ofm. Thenr(a) = (Pl"'PT) = nl~l(pJ

Proposition 1.16. Let a, b be ideals in a ring A such that r(a), reb) are coprime. Then a, b are coprime.

Proof rea + b) = r(r(a) + reb»~ = r(I) = (1), hence a + b = (1) by (1.13).

EXTENSION AND CONTRACTION

Letf: A --+ B be a ring homomorphism. If a is an ideal in A, the setf(a) is not necessarily an ideal in B (e.g., let f be the embedding of Z in Q, the field of rationals, and take a to be any nOn-zero ideal in Z.) We define the extension ae of a to be the ideal Bf(a) generated by f(a) in B: explicitly, ae is the set of all sums 2: yJ(Xi) where Xi E a, Yi E B.

If b is an ideal of B, then f-l(b) is always an ideal of A, called the contrac­tion bC orb. 1[0 is prime, then bC is prime. If a is prime, ae need not be prime (for example, f: Z --+ Q, a t= 0; then ae = Q, which is not a prime ideal).

We can factorize f as follows:

A ~f(A) ~ B

where p is surjective and j is injective. For p the situation is very simple (1.1): there is a one-to-one correspondence between ideals of f(A) and ideals of A which contain Ker (f), and prime ideals correspond to prime ideals. For j, on the other hand, the general situation is very complicated. The classical example is from algebraic number theory.

10 RINGS AND IDEALS

Example. Consider Z -+ Z[i], where i = v=1. A prime ideal (p) of Z mayor may not stay prime when extended to Z[i]. In fact Z[i] is a principal ideal domain (because it has a Euclidean algorithm) and the situation is as follows:

i) (2)" = (1 + i)2), the square of a prime ideal in Z[i];

ii) If p == 1 (mod 4) then (p)e is the product of two distinct prime ideals (for example, (5)e = (2 + i)(2 - i));

iii) If p == 3 (mod 4) then (p)e is prime in Z[i].

Of these, ii) is not a trivial result. It is effectively equivalent to a theorem of Fermat which says that a prime p == 1 (mod 4) can be expressed, essentially uniquely, as a sum of two integer squares (thus 5 = 22 + 12,97 = 92 + 42, etc.).

In fact the behavior of prime ideals under extensions of this sort is one of the central problems of algebraic number theory.

Let f: A -+ B, a and b be as before. Then

Proposition 1.17. i) a s; aec, b ;:> bC';

ii) bC = bC'

c, ae = aece ;

iii) If C is the set of contracted ideals in A and if E is the set of extended ideals in B, then C = {a/ aee = a}, E = {b / bee = b}, and a f-+ ae is a bijective map of C onto E, whose inverse is b f-+ be.

Proof i) is trivial, and ii) follows from i).

iii) If a E C, then a = be. = bcee = aee ; conversely if a = aee then a is the contraction of ae• Similarly for E. •

Exercise 1.18. If alo a2 are ideals of A and ifb!> b2 are ideals of B, then

(al + a2 )" = a~ + a~, (al n a2)e s; a~ (") a~, (a1a2), = a~a~,

(al : a2)e s; (a~: ~), rea)" s; r(ae),

(b 1 + b2)" ;:> bf + b~, (b1 n b2)" = b~ n b~, (b 1b2)" ;:> b~b~,

(b 1 :b2)" s; (b~:b~), reb y = r(b C

).

The set of ideals E is closed under sum and product, and C is closed under the other three operations.

EXERCISES

1. Let x be a nilpotent element of a ring A. Show that 1 + x is a unit of A. Deduce that the sum of a nilpotent element and a unit is a unit.

2. Let A be a ring and let A[x] be the ring of polynomials in an indeterminate x, with coefficients in A. Let f = ao + alX + ... + anxn E A[x]. Prove that

EXERCISES II

i) I is a unit in A (x] ..... ao is a unit in A and al, ... , an are nilpotent. [If bo + b1x + ... + bmxm is the inverse of J. prove by induction on r that a~+lbm_r = O. Hence show that an is nilpotent, and then use Ex. 1.]

ii) I is nilpotent -= ao, al, ... , an are nilpotent. iii) I is a zero-divisor -= there exists a -:f. 0 in A such that al = O. [Choose a

polynomial g = bo + b1x + ... + bmxm of least degree m such thatlg = O. Then anbm = 0, hence ang = 0 (because ang annihilates I and has degree < m). Now show by induction that an-rg = 0 (0 ~ r ~ n).]

iv) I is said to be primitive if (ao, aI, ... , an) = (1). Prove that if J. g E A (x], then Ig is primitive ¢> I and g are primitive.

3. Generalize the results of Exercise 2 to a polynomial ring A(Xl, ... , xrJ in several indeterminates.

4. In the ring A(x), the Jacobson radical is equal to the nilradical.

S. Let A be a ring and let A((xJ] be the ring offormal power series I = :L.:.o a,.xn with coefficients in A. Show that

i) lis a unit in A((x)) ¢> ao is a unit in A. ii) If I is nilpotent, then a" is nilpotent for all n ~ O. Is the converse tlue?

(See Chapter 7, Exercise 2.) iii) I belongs to the Jacobson radical of A((x]J -= ao belongs to the Jacobson

radical of A. iv) The contraction of a maximal ideal m of A((x]J is a maximal ideal of A, and

m is generated by me and x. v) Every prime ideal of A is the contraction of a prime ideal of A [[xJ].

6. A ring A is such that every ideal not contained in the nilradical contains a non­Zero idempotent (that is, an element e such that e2 = e -:f. 0). Prove that the nilradical and Jacobson radical of A are equal.

7. Let A be a ring in which every element x satisfies x" = x for some n > 1 (depending on x). Show that every prime ideal in A is maximal.

8. Let A be a ring -:f. O. Show that the set of prime ideals of A has minimal ele­ments with respect to inclusion.

9. Let a be an ideal -:f. (1) in a ring A. Show that a = rea) - a is an intersection of prime ideals.

10. Let A be a ring, 91 its nilradical. Show that the following are equivalent: i) A has exactly one prime ideal; ii) every element of A is either a unit or nilpotent;

iii) Aj91 is a field.

11. A ring A is Boolean if x 2 = x for all x E A. In a Boolean ring A, show that i) 2x = 0 for all x E A ; ii) every prime ideal p is maximal, and A/p is a field with two elements;

iii) every finitely generated ideal in A is principal.

12. A local ring contains no idempotent -:f. 0, 1.

Construction 01 an algebraic closure 01 a field (E. Artin). 13. Let K be a field and let ~ be the set of all irreducible monic polynomials lin one

12 RINGS AND IDEALS

indeterminate with coefficients in K. Let A be the polynomial ring over K generated by indeterminates x" one for each /E 1:. Let a be the ideal of A generated by the polynomials/ex,) fgr all/E 1:. Show that a :j:. (1).

Let m be a maximal ideal of A containing a, and let Kl = Aim. Then Kl is an extension field of K in which each IE 1: has a root. Repeat the construction with Kl in place of K, obtaining a field K2 , and so on. Let L = U:=l Kn. Then L is a field in which each/E 1: splits completely into linear factors. Let K be the set of all elements of L which are algebraic over K. Then K is an algebraic closure of K.

14. In a ring A, let 1: be the set of all ideals in which every element is a zero-divisor. Show that the set 1: has maximal elements and that every maximal element of 1: is a prime ideal. Hence the set of zero-divisors in A is a union of prime ideals.

The prime spectrum 0/ a ring

15. Let A be a ring and let X be the set {)f all prime ideals of A. For each subset E of A, let VeE) denote the set of all prime ideals of A which contain E. Prove that

i) if a is the ideal generated by E, then VeE) = V(a) = V(r(a»). ii) V(O) = X, V(1) = 0.

iii) if (E,),el is any family of subsets of A, then

V( U E,) = n VeE,). ,el ,el

iv) V(a (\ a) = V(aa) = V(a) U V(b) for any ideals a, a of A. These results show that the sets VeE) satisfy the axioms for clo.sed sets

in a topological space. The resulting topology is called the Zariski topology. The topological space X is called the prime spectrum of A, and is written Spec (A).

16. Draw pictures of Spec (Z), Spec (R), Spec (C[x]), Spec (R[x]), Spec (~[x])_

17. For each /E A, let X, denote the complement of V(f) in X = Spec (A). The sets X, are open. Show that they form a basis of open sets for the Zariski topology, and that

i) X, (\ Xg = XIg ; ii) X, = 0 -= I is nilpotent;

iii) XI = X -= lis a unit; iv) XI = Xg -= r((I») = r(g»); v) X is quasi-compact (that is, every open covering of X has a finite sub­

covering). vi) More generally, each XI is quasi-compact. vii) An open subset of X is quasi-compact if and only if it is a finite union of

sets XI' The sets XI are called basic open sets of X = Spec (A).

[To prove (v), remark that it is enough to consider a covering of X by basic open sets XI, (i E I). Show that theft generate the unit ideal and hence that there is an equation of the form

(g, E A)

where J is some finite su bset of I. Then the Xli (i E J) cover X.]

EXERCISES 13

18. For psychological reasons it is sometimes convenient to denote a prime ideal of A by a letter such as x or y when thinking of it as a point of X = Spec (A). When thinking of x as a prime ideal of A, we denote it by tJx (logically, of course, it is the same thing). Show that

i) the set {x} is closed (we say that x is a "closed point") in Spec (A) «> tJx is maximal;

ii) {x} = V(lJx);

iii) yE{X} «>tJx S lJy; iv) X is a To-space (this means that if x, yare distinct points of X, then either

there is a neighborhood of x which does not contain y, or else there is a neighborhood of y which does not contain x).

19. A topological space X is said to be irreducible if X :j:. 0 and if every pair of non-empty open sets in X intersect, or equivalently if every non-empty open set is dense in X. Show that Spec (A) is irreducible if and only if the nilradical of A is a prime ideal.

20. Let X be a topological space. i) If Y is an irreducible (Exercise 19) subspace of X, then the closure Yof Y

in X is irreducible. ii) Every irreducible subspace of X is contained in a maximal irreducible

subspace. iii) The maximal irreducible subspaces of X are closed and cover X. They are

called the irreducible components of X. What are the irreducible components of a Hausdorff space?

iv) If A is a ring and X = Spec (A), then the irreducible components of X are the closed sets V(tJ), where lJ is a minimal prime ideal of A (Exercise 8).

21. Let </>: A ---+ B be a ring homomorphism. Let X = Spec (A) and Y = Spec (B). If q E Y, then </> - l(q) is a prime ideal of A, i.e., a point of X. Hence </> induces a mapping </>*: Y -+ X. Show that

i) If IE A then </>*-l(X,) = Y<t><th and hence that </>* is continuous. ii) If 0 is an ideal of A, then </>*-l(V(O) = V(oe).

iii) If b is an ideal of B, then </>*(V(b) = VW). iv) If </> is surjective, then </>* is a homeomorphism of Yonto the closed subset

V(Ker (</») of X. (In particular, Spec (A) and Spec (A/'R) (where 'R is the nilradical of A) are naturally homeomorphic.)

v) If </> is injective, then </>*( Y) is dense in X. More precisely, </>*( Y) is dense in X «> Ker (</» S 'R.

vi) Let.p: B -+ C be another ring homomorphism. Then (.p 0 </»* = </>* o.p*. vii) Let A be an integral domain with just one non-zero prime ideallJ, and let K

be the field of fractions of A. Let B = (A/lJ) x K. Define </>: A -+ B by </>(x) = (x, x), where x is the image of x in A/tJ. Show that </>* is bijective but not a homeomorphism.

22. Let A = TIf = 1 A, be the direct product of rings A,. Show that Spec (A) is the disjoint union of open (and closed) subspaces X" where X, is canonically homeomorphic with Spec (A,).

14 RINGS AND IDEALS

Conversely, let A be any ring. Show that the following statements are equivalent:

i) X = Spec (A) is disconnected. ii) A ~ Al X A2 where neither of the rings A1 , A2 is the zero ring.

iii) A contains an idempotent #: 0, 1. In particular, the spectrum of a local ring is always connected (Exercise

12).

23. Let A be a Boolean ring (Exercise 11), and let X = Spec (A). i) For each/E A, the set XI (Exercise 17) is both open and closed in X.

ii) Let /1, .. . .In E A. Show that Xfl u· .. U XI" = XI for some /E A. iii) The sets XI are the only subsets of X which are both open and closed.

[Let Y S X be both open and closed. Since Y is open, it is a union of basic open sets XI' Since Y is closed and X is quasi-compact (Exercise 17), Y is quasi-compact. Hence Y is a finite union of basic open sets; now use (ii) above.]

iv) X is a compact Hausdorff space.

24. Let L be a lattice, in which the sup and inf of two elements a, b are denoted by a V b and a 1\ b respectively. L is a Boolean lattice (or Boolean algebra) if

i) L has a least element and a greatest element (denoted by 0, 1 respectively). ii) Each of v, 1\ is distributive over the other. iii) Each a E L has a unique "complement" a' E L such that a V a' = 1 and

a 1\ a' = O. (For example, the set of all subsets of a set, ordered by inclusion, is a Boolean lattice.)

Let L be a Boolean lattice. Define addition and multiplication in L by the rules

a + b = (a 1\ b') V (a' 1\ b), ab = a 1\ b.

Verify that in this way L becomes a Boolean ring, say A(L). Conversely, starting from a Boolean ring A, define an ordering on A as

follows: a .::; b means that a = abo Show that, with respect to this ordering, A is a Boolean lattice. [The sup and inf are given by a V b = a + b + ab and a 1\ b = ab, and the complement by a' = 1 - a.] In this way we obtain a one-to-one correspondence between (isomorphism classes of) Boolean rings and (isomorphism classes of) Boolean lattices.

25. From the last two exercises deduce Stone's theorem, that every Boolean lattice is isomorphic to the lattice of open-and-closed subsets of some compact Haus­dorff topological space.

26. Let A be a ring. The subspace of Spec (A) consisting of the maximal ideals of A, with the induced topology, is called the maximal spectrum of A and is denoted by Max (A). For arbitrary commutative rings it does not have the nice functorial properties of Spec (A) (see Exercise 21), because the inverse image of a maximal ideal under a ring homomorphism need not be maximal.

Let X be a compact Hausdorff space and let C(X) denote the ring of all real-valued continuous functions on X (add and multiply functions by adding

EXERCISES 15

and multiplying their values). For each x E X, let mx be the set of all IE C(X) such that I(x) = O. The ideal mx is maximal, because it is the kernel of the (surjective) homomorphism C(X) -- R which takes I to I(x). If g denotes Max (C(X», we have therefore defined a mapping p.: X -+ g, namely x!-+ m x •

We shall show that p. is a homeomorphism of X onto g. i) Let m be any maximal ideal of C(X), and let V = V(m) be the set of com­

mon zeros of the functions in m: that is,

V = {x E X:/(x) = 0 for all/E m}.

Suppose that V is empty. Then for each x E X there exists Ix Em such that lAx) ¥; O. Since Ix is continuous, there is an open neighborhood Ux of x in X on which Ix does not vanish. By compactness a finite number of the neighborhoods, say Ux;, ... , Ux., cover X. Let

I = 1;1 + ... + Ix~' Then I does not vanish at any point of X, hence is a unit in C(X). But this contradicts IE m, hence V is not empty.

Let x be a point of V. Then m s m x , hence m = mx because m is maximal. Hence p. is surjective.

ii) By Urysohn's lemma (this is the only non-trivial fact required in the argu­ment) the continuous functions separate the points of X. Hence x ¥; y =>

mx ¥; my, and therefore p. is injective. iii) Let IE C(X); let

Ur = {x E X:/(x) ¥; O}

and let Ur = {m E X:f¢ m}

Show that ,.,.(Ur) = Ur. The open sets U, (resp. U,) form a basis of the top­ology of X (resp. g) and therefore p. is a homeomorphism. Thus X can be reconstructed from the ring of functions C(X).

Affine algebraic varieties

27. Let k be an algebraically closed field and let

laCt!,"" tn) = 0

be a set of polynomial equations in n variables with coefficients in k. The set X of all points x = (Xl, ... , x n ) E k n which satisfy these equations is an affine algebraic variety.

Consider the set of all polynomials g E k[tl, ... , In] with the property that g(x) = 0 for all x E X. This set is an ideal I(X) in the polynomial ring, and is called the ideal 01 the variety X. The quotient ring

P(X) = k[/ l , . .. , tn]/I(X)

is the ring of polynomial functions on X, because two polynomials g, h define the same polynomial function on X if and only if g - Ir vanishes at every point of X, that is, if and only if g - /z E leX).

16 RINGS AND IDEALS

Let e, be the image of t, in P(X). The e, (1 :os; i :os; n) are the coordinate functions on X: if x E X, then Ux) is the ith coordinate of x. P(X) is generated as a k-algebra by the coordinate functions, and is called the coordinate ring (or affine algebra) of X.

As in Exercise 26, for each x E X let mx be the ideal of allfE P(X) such that f(x) = 0; it is a maximal ideal of P(X). Hence, if X = Max (P(X)), we have defined a mapping f.L: X -+ X, namely x I-> m x .

It is easy to show that f.L is injective: if x ¥; y, we must have x, ¥; y, for for some i (1 :os; i :os; n), and hence e, - x, is in mx but not in my, so that mx ¥; my. What is less obvious (but still true) is that f.L is surjective. This is one form of Hilbert's Nullstellensatz (see Chapter 7).

28. Letfh'" ,fm be elements of k[tl , .. . , tnl. They determine a polynomial mapping 4>: kn -+ k m: if x E kn, the coordinates of 4>(x) are fleX), ... .Jm(X).

Let X, Y be affine algebraic varieties in kn, km respectively. A mapping 4>: X -+ Yis said to be regular if 4> is the restriction to X of a polynomial map­ping from kn to km.

If '7 is a polynomial function on Y, then '7 0 4> is a polynomial function on X. Hence 4> induces a k-algebra homomorphism P( Y) -+ P(X), namely '7 I-> '7 0 rp. Show that in this way we obtain a one-to-one correspondence between the regular mappings X -+ Y and the k-algebra homomorphisms P( Y) -+ P(X).

z

Modules

One of the things which distinguishes the modern approach to Commutative Algebra is the greater emphasis on modules, rather than just on ideals. The extra "elbow-room" that this gives makes for greater clarity and simplicity. For instance, an ideal a and its quotient ring A/a are both examples of modules and so, to a certain extent, can be treated on an equal footing. In this chapter we give the definition and elementary properties of modules. We also give a brief treatment of tensor products, including a discussion of how they behave for exact sequences.

MODULES AND MODULE HOMOMORPHISMS

Let A be a ring (commutative, as always). An A-module is an abelian group M (written additively) on which A acts linearly: more precisely, it is a pair (M,IL), where M is an abelian group and IL is a mapping of A x Minto M such that, if we write ax for IL(a, x)(a E A, x EM), the following axioms are satisfied:

a(x + y) = ax + ay, (a + b)x = ax + bx,

(ab)x = a(bx), Ix = x Ca, b E A; x, Y EM).

(Equivalently, M is an abelian group together with a ring homomorphism A -+ E(M), where E(M) is the ring of endomorphisms of the abelian group M.)

The notion of a module is a common generalization of several familiar concepts, as the following examples show:

Examples. 1) An ideal a of A is an A-module. In particular A itself is an A-module.

2) If A is a field k, then A-module = k-vector space.

3) A = Z, then Z-module = abelian group (define nx to be x+ ... + x).

4) A = k[x] where k is a field; an A-module is a k-vector space with a linear transformation.

5) G = finite group, A = k[G] = group-algebra of G over the field k (thUS A is not commutative, unless G is). Then A-module = k-representation of G.

17

18 MODULES

Let M, N be A-modules. A mapping f: M ~ N is an A-module homo­morphism (or is A-linear) if

f(x + y) = f(x) + fey) f(ax) = aI(x)

for all a E A and all x, y E M. Thus f is a homomorphism of abelian groups which commutes with the action of each a EA. If A is a field, an A-module homomorphism is the same thing as a linear transformation of vector spaces.

The composition of A-module homomorphisms is again an A-module homomorphism.

The set of all A-module homomorphisms from M to N can be turned into an A-module as follows: we define f + g and af by the rules

(f + g)(x) = f(x) + g(x), (af)(x) = aI(x)

for all x E M. It is a trivial matter to check that the axioms for an A-module are satisfied. This A-module is denoted by HomA (M. N) (or just Hom (M, N) if there is no ambiguity about the ring A).

Homomorphisms u: M' ~ M and v: N ~ N" induce mappings

u: Hom (M, N) ~ Hom (M', N) and v: Hom (M, N) ~ Hom (M, N")

defined as follows:

u(f) = f 0 u, v(f) = v 0 f These mappings ~re A-module homomorphisms.

For any module M there is a natural isomorphism Hom (A, M) ~ M: any A-module homomorphism f: A ~ M is uniquely determined by f{l), which can be any element of M.

SUBMODULES AND QUOTIENT MODULES

A submodule M' of M is a subgroup of M which is closed under multiplication by elements of A. The abelian group MIM' then inherits an A-module structure from M, defined by a(x + M') = ax + M'. The A-module MIM' is the quotient of M by M'. The natural map of M onto MIM' is an A-module homo­morphism. There is a one-to-one order-preserving correspondence between submodules of M which contain M', and submodules of M" Uust as for ideals; the statement for ideals is a special case).

Iff: M ~ N is an A-module homomorphism, the kernel offis the set

Ker (f) = {x E M:f(x) = O}

and is a submodule of M. The image off is the set

1m (f) = f(M)

OPERATIONS ON SUBMODULES 19

and is a submodule of N. The cokernel off is

Coker (f) = N/Im (J)

which is a quotient module of N. If M' is a submodule of M such that M' ~ Ker (f), thenfgives rise to a

homomorphism]: M/M' -+ N, defined as follows: if x E M/M' is the image of x EM; then](x) = f(x). The kernel of]is Ker (f)/M'. The homomorphism] is said to be induced by f In particular, taking M' = Ker (I), we have an isomorphism of A-modules

M/Ker (f) ~ 1m (I).

OPERATIONS ON SUBMODULES

Most of the operations on ideals considered in Chapter 1 have their counter­parts for modules. Let M be an A-module and let (M')lel be a family of sub­modules of M. Their sum L M, is the set of all (finite) sumS L x" where x, E M, for all i E I, and almost all the Xi (that is, all but a finite number) are zero. L M, is the smallest submodule of M which contains all the MI'

The intersection n Ml is again a submodule of M. Thus the submodules of M form a complete lattice with respect to inclusion.

Proposition 2.1. i) If L :2 M :2 N are A-modules, then

(L/ N)/(M/ N) ~ L/ M.

ii) If Mh M2 are submodules of M, then

(Ml + M 2)/M1 ~ M 2/(M1 (') M2)'

Proof i) Define 0: L/ N -+ L/ M by O(x + N) = x + M. Then 0 is a well­defined A-module homomorphism of L/N onto L/M, and its kernel is M/N; hence (i).

ii) The composite homomorphism M2 -+ Ml + M2 -+ (Ml + M 2)/M1 is surjective, and its kernel is Ml (') M 2 ; hence (ii). •

We cannot in general define the product of two submodules, but we can define the product aM, where a is an ideal and M an A-module; it is the set of all finite sums 2: alXI with a, E a, Xl E M, and is a submodule of M.

If N, Pare submodules of M, we define (N:P) to be the set of all a E A such that aP ~ N; it is an ideal of A. In particular, (0: M) is the set of all a E A such that aM = 0; this ideal is called the annihilator of M and is also denoted by Ann (M). If a ~ Ann (M), we may regard M as an A/a-module, as follows: if x E A/a is represented by X E A, define xm to be xm(m E M): this is independ­ent of the choice of the representative x of x, since aM = O.

20 MODULES

An A-module is faithful if Ann (M)=O. If Ann (M) = a, then M is faithful as an A/a-module.

Exercise 2.2. i) Ann (M + N) = Ann (M) (') Ann (N).

ii) (N:P) = Ann «N + P)/N).

If x is an element of M, the set of all multiples ax(a E A) is a submodule of M, denoted by Ax or (x). If M = LjeI AXh the Xj are said to be a set of gen­erators of M; this means that every element of M can be expressed (not neces­sarily uniquely) as a finite linear combination of the Xj with coefficients in A. An A-module M is said to be finitely generated if it has a finite set of genera­tors.

DIRECT SUM AND PRODUCT

If M, N are A-modules, their direct sum M EB N is the set of all pairs (x, y) with x E M, YEN. This is an A-module if we define addition and scalar multiplica­tion in the obvious way:

(Xl' Yl) + (X2' Y2) = (Xl + X2, Yl + Y2) a(x, y) = (ax, ay).

More generally, if (Mj)jeI is any family of A-modules, we can define their direct sum EE\eI M j ; its elements are families (Xi)ieI such that Xj E Mi for each i E I and almost all XI are 0. If we drop the restriction on the number of non-zero x's we have the direct product OjeI Mi' Direct sum and direct product are therefore the same if the index set I is finite, but not otherwise, in general.

Suppose that the ring A is'a direct product Of "'I Ai (Chapter 1). Then the set of all elements of A of the form

(0, ... , 0, at. 0, ... , 0)

with aj E Ai is an ideal OJ of A (it is not a subring of A-except in trivial cases­because it does not contain the identity element of A). The ring A, considered as an A-module, is the direct sum of the ideals 01> ... , an' Conversely, given a module decomposition

of A as a direct sum of ideals, we have

where OJ = EBi"t OJ. Each ideal t1j is a ring (isomorphic to A/Oj). The identity element ei of OJ is an idempotent in A, and ai = (ei). ..

FlNITEL Y GENERATED MODULES 21

FINITELY GENERATED MODULES

A free A-module is one which is isomorphic to an A-module of the form EBlel Mh where each Ml ~ A (as an A-module). The notation A(I) is sometimes used. A finitely generated free A-module is therefore isomorphic to A EB· .. EEl A (n summands), which is denoted by An. (Conventionally, AO is the zero module, denoted by 0.)

Proposition 2.3. M is a finitely generated A-module ~ M is isomorphic to a quotient of An for some integer n > O.

Proof =>: Let Xl> .•. , Xn generate M. Define ep: An --+ M by ep(alo ... , an) = alx1 + ... + anxn. Then ep is an A-module homomorphism onto M, and there­fore M ~ An/Ker (ep).

<:=: We have an A-module homomorphism ep of An onto M. If ej = (0, ... ,0, 1,0, ... ,0) (the 1 being in the ith place), then the el (1 ~ i ~ n) generate An, hence the ep(ej) generate M. _

Proposition 2.4. Let M be afinitely generated A-module, let a be an ideal of A, and let ep be an A-module endomorphism of M such that ep(M) S; a M. Then ep satisfies an equation of the form

where the aj are in a.

Proof Let Xl> ..• , Xn be a set of generators of M. Then each ep(XI) E aM, so that we have say ep(Xj) = 'Lf = 1 ajjXj (1 ~ i ~ n; ali E a), i.e.,

n

2: (aIJ<P - alj)Xj = 0 j= 1

where alj is the Kronecker delta. By multiplying on the left by the adjoint of the matrix (aliep - aij) it follows that det (ajjep - ali) annihilates each Xl' hence is the zero endomorphism of M. Expanding out the determinant, we have an equation of the required form. _

Corollary 2.5. Let M be a finitely generated A-module and let a be an ideal of A such that aM = M. Then there exists X == l(mod a) such that xM = O.

Proof Take ep = identity, X = 1 + a l + ... + an in (2.4). _

Proposition 2.6. (Nakayama's lemma). Let M be a finitely generated A-module and a an ideal of A contained in the Jacobson radical 9t of A. Then aM = M implies M = O.

Fir~t Proof By (2.5) we have xM = 0 for some x == 1 (mod 9t). By (1.9) X is a unit in A, hence M = x-1xM = O. _

22 MODULES

Second Proof. Suppose M =F 0, and let Ul> ... , Un be a minimal set of gener­ators of M. Then Un E aM, hence we have an equation of the form Un = a1U1 + ... + anun, with the a, E a. Hence

(l - an)un = a1U1 + ... + an- 1Un-1;

since an E St, it follows from (1.9) that 1 - an is a unit in A. Hence Un belongs to the submodule of M generated by Ul> ... , Un -1: contradiction. -

Corollary 2.7. Let M be afinitely generated A-module, N a submodule of M, a ~ St an ideal. Then M = aM + N => M = N.

Proof. Apply (2.6) to MIN, observing that a(MIN) = (aM + N)IN. _

Let A be a local ring, m its maximal ideal, k = Aim its residue field. Let M be a finitely generated A-module. MlmM is annihilated by m, hence is naturally an Aim-module, i.e., a k-vector space, and as such is finite-dimensional.

Proposition 2.8. Let x, (1 ~ i ~ n) be elements of M whose images in MlmM form a basis of this vector space. Then the x, generate M.

Proof. Let N be the submodule of M generated by the X,. Then the composite map N --+ M --+ MlmM maps N onto MlmM, hence N + mM = M, hence N = M by (2.7). _

EXACf SEQUENCES

A sequence of A-modules and A-homomorphisms

M Il M 11+1 M ... ~ '-1- ,~ '+1-'" (0)

is said to be exact at M, if 1m (J;) = Ker (J; + 1)' The sequence is exact if it is exact at each M,. In particular:

o --+ M' .4. M is exact <:> f is injective; (I)

M ~ M" --+ 0 is exact <:> g is surjective; (2)

0--+ M' .4. M 4 M" --+ 0 is exact <:> f is injective, g is surjective and g induces an isomorphism of Coker if) = Mlf(M') onto M". (3)

A sequence of type (3) is called a short exact sequence. Any long exact sequence (0) can be split up into short exact sequences: if N, = 1m (J;) =

Ker (J; + 1), we have short exact sequences 0 --+ N, --+ M, --+ N, + 1 --+ 0 for each i.

Proposition 2.9. i) Let

M' ~ M 4 M" --+ 0 (4)

be a sequence of A-modules and homomorphisms. Then the sequence (4) is exact <=> for all A-modules N, the sequence

0--+ Hom (M", N) ~ Hom (M, N) ~ Hom (M', N) (4')

is exact.

EXACT SEQUENCES 23

ii) Let

(5)

be a sequence of A~modules and homomorphisms. Then the sequence (5) is exact <=> for all A~modules M, the sequence

0--+ Hom (M, N') ~ Hom (M, N) ~ Hom (M, N") (5')

is exact.

All four parts of this proposition are easy exercises. For example, suppose that (4') is exact for all N. First of all, since v is injective for all N it follows that v is surjective. Next, we have u 0 v = 0, that is v 0 u 0 f = 0 for allf: M" --+- N. Taking N to be M" andfto be the identity mapping, it follows that v 0 u = 0, hence 1m (u) £, Ker (v). Next take N = MJlm (u) and let </>: M --+- N be the projection. Then cp E Ker (u), hence there exists 1/1: M" -+ N such that </> = 1/1 0 v. Consequently 1m (u) = Ker (</» ;2 Ker (v). _

Proposition 2.10. Let

o --+- M' ~ M 4 M" -+ 0

0-+ N' ? N? N" -+ 0

be a commutative diagram of A-modules and homomorphisms, with the rows exact. Then there exists an exact sequence

o -+ Ker (f') ~ Ker if) ~ Ker (f") ~ Q' ~

Coker (f') -- Coker if) -- Coker (f") --,). 0 (6)

in which ii, v are restrictions of u, v, and ii', v' are induced by u', v'.

The boundary homomorphism d is defined as follows: if x· E Ker (f"), we have x" = vex) for some x E M, and v'(j(x») = j"(v(x» = 0, hencef(x) E Ker (v') = 1m (u'), so that f(x) = u'(y') for some y' EN'. Then d(x") is defined to be the image of y' in Coker (f'). The verification that d is well-defined, and that the sequence (6) is exact, is a straightforward exercise in diagram-chasing which we leave to the reader. _

Remark. (2.10) is a special case of the exact homology sequence of homological algebra.

Let C be a class of A-modules and let A be a function on C with values in Z (or, more generally, with values in an abelian group 0. The function A is additive if, for each short exact sequence (3) in which all the terms belong to C, we have A(M') - '\(M) + A(M") = O.

Example. Let A be a field k, and let C be the class of all finite-dimensional k-vector spaces V. Then V ~ dim V is an additive function on C.

2+r.C.A.

24 MODULES

Proposition 2.11. Let 0 ~ Mo ~ Ml~"'~ Mn ~ 0 be an exact se­quence of A-modules in which all the modules M, and the kernels of all the homomorphisms belong to C. Then for any additive function A on C we have

n

L (-l)lA(M,) = O. '=0

Proof Split up the sequence into short exact sequences

o ~ N j ~ oM, ~ N j + 1 ~ 0

(No = N"+l = 0). Then we have A(M,) = A(N,) + A(Nj+l)' Now take the alternating sum of the A(M,), and everything cancels out. •

TENSOR PRODUcr OF MODULES

Let M, N, P be three A-modules. A mapping f: M x N ~ P is said to be A-bilinear if for each x EM the mappingy 1--+ f(x, y) of N into P is A-linear, and for each yEN the mapping x 1--+ f(x, y) of Minto P is A-linear.

We shall construct an A-module T, called the tensor product of M and N, with the property that the A-bilinear mappings M x N ~ P are in a natural one-to-one correspondence with the A-linear mappings T ~ P, for all A­modules P. More precisely:

Proposition 2.12. Let M, N be A-modules. Then there exists a pair (T, g) consisting of an A-module T and an A-bilinear mapping g: M x N ~ T, with the following property:

Given any A-module P and any A-bilinear mapping f: M x N ~ P, there exists a unique A-linear mapping f': T ~ P such that f = f' 0 g (in other words, every bilinear function on M x N factors through T).

Moreover, if(T, g) and (T', g') are two pairs with this property, then there exists a unique isomorphism j: T ~ T' such that jog = g'.

Proof i) Uniqueness. Replacing (P,!) by (T', g') we get a unique j: T ~ T' such that g' = jog. Interchanging the roles of Tand T', we getj': T' ~ Tsuch that g = j' 0 g'. Each of the compositionsj 0 j',j' 0 j must be the identity, and therefore j is an isomorphism.

ii) Existence. Let C denote the free A-module A(MXN). The elements of C are formal linear combinations of elements of M x N with coefficients in A, i.e. they are expressions of the form 2f= 1 aj' (x" Yj)(a, E A, x, E M, y, EN).

Let D be the submodule of C generated by all elements of C of the follow­ing types:

(x + x', y) - (x, y) - (x', y) (x, y + y') - (x, y) - (x, y')

(ax, y) - a·(x, y) (x,ay) - a·(x,y).

TENSOR PRODUCT OF MODULES 25

Let T = C/ D. For each basis element (x, y) of C, let x ® y denote its image in T. Then T is generated by the elements of the form x ® y, and from our definitions we have

(x + x') 181 y = x 181 y + x' 181 y, x 181 (y + y') = x 181 y + x ® y', (ax) 181 y = x 181 (ay) = a(x 181 y)

Equivalently, the mapping g: M x N -+ T defined by g(x, y) = x ® y is A-bilinear.

Any map f of M x N into an A-module P extends by linearity to an A­module homomorphism 1: C -+ P. Suppose in particular that f is A-bilinear. Then, from the definitions, 1 vanishes on all the generators of D, hence on the whole of D, and therefore induces a well-defined A-homomorphism f' of T = C/D into P such that f'(x 181 y) = f(X, y). The mapping f' is uniquely defined by this condition, and therefore the pair (T, g) satisfy the conditions of the proposition. _

Remarks. i) The module T constructed above is called the tensor product of M and N, and is denoted by M I8I A N, or just M 181 N if there is no ambiguity about the ring A. It is generated as an A-module by the "products" x 181 y. If (X')iel> (Yf)fel are families of generators of M, N respectively, then the elements XI 181 YJ generate M ® N. In particular, if M and N are finitely generated, so is M 181 N.

ii) The notation x 181 y is inherently ambiguous unless we specify the tensor product to which it belongs. Let M', N' be submodules of M, N respectively, and let x EM' and YEN'. Then it can happen that x 181 y as an element of M ® N is zero whilst x ® y as an element of M' ® N' is non-zero. For example, take A = Z, M = Z, N = Z/2Z, and let M' be the submodule 2Z of Z, whilst N' = N. Let x be the non-zero element of N and consider 2 181 x. As an element of M 181 N, it is zero because 2 181 x = 1 181 2x = 1 181 0 = O. But as an element of M' 181 N' it is non-zero. See the example after (2.18).

However, there is the following result:

Corollary 2.13. Let Xl E M, y, E N be such that L x, 181 y, = 0 in M ® N. Then there existfinitely generated submodules Mo of M and No of N such that L x, ® y, = 0 in Mo 181 No·

Proof If LX, 181 y, = 0 in M 181 N, then in the notation of the proof of (2.11.) we have L (x" y,) E D, and therefore L (x" y,) is a finite sum of generators of D. Let Mo be the submodule of M generated by the Xi and all the elements of M which occur as first coordinates in these generators of D, and define No simi­larly. Then LX, 181 y, = 0 as an element of Mo 181 No. -

iii) We shall never again need to USe the construction of the tensor product given in (2.12), and the reader may safely forget it if he prefers. What is essential to keep in mind is the defining property of the tensor product.

26 MODULES

iv) Instead of starting with bilinear mappings we could have started with multilinear mappings f: Ml x··· X Mr ~ P defined in the same way (i.e., linear in each variable). Following through the proof of (2.12) we should end up with a "multi-tensor product" T = Ml 18i ... ® M" generated by all products Xl ® ... ® X T (XI E Mj) 1 ~ i ~ r). The details may safely be left to the reader; the result corresponding to (2.12) is

Proposition 2.12*. Let M lo . •. , Mr be A-modules. Then there exists a pair (T, g) consisting of an A-module T and an A-multilinear mapping g: Ml x··· X Mr ~ T with the following property:

Given any A-module P and any A-multilinear mapping f: Ml x··· X MT ~ T, there exists a unique A-homomorphism f': T ~ P such that f'og=f.

Moreover, if(T, g) and (T', g') are two pairs with this property, then there exists a unique isomorphism j: T --+ T' such that jog = g'. •

There are various so-called "canonical isomorphisms", some of which we state here:

Proposition 2.14. Let M, N, P be A-modules. Then there exist unique isomorphisms

i)M®N~N®M

ii) (M ® N) ® P ~ M I8i (N I8i P) ~ M I8i N I8i P

iii) (M EB N) ® P ~ (M ® P) EEl (N ® P)

iv) A ® M~M

such that, respectively,

a) x ® y f-+ Y ® x

b) (x ® y) ® Z f-+ X ® (y ® z) 1-+ X I8i y I8i Z

c) (x, y) ® Z f-+ (x ® z, y ® z)

d) a ® x f-+ ax. Proof. In each case the point is to show that the mappings so described are well defined. The technique is to construct suitable bilinear or multilinear mappings, and use the defining property (2.12) or (2.12*) to infer the existence of homo­morphisms of tensor products. We shall prove half of ii) as an example of the method, and leave the rest to the reader.

We shall construct homomorphisms

(M ® N) ® P 4. M I8i N ® P ~ (M I8i N) ® P

such thatf«x ® y) ® z) = x ® y I8i Z and g(x ® y I8i z) = (x ® y) ® z for all x E M, YEN, Z E P.

To construct J, fix the element Z E P. The mapping (x, y) f-+ X ® y ® Z

(x E M, YEN) is bilinear in x and y and therefore induces a homomorphism

RESTRICTION AND EXTENSION OF SCALARS 27

fz: M ® N ~ M I8i N ® P such thatfz(x ® y) = x ® y I8i z. Next, consider the mapping (t, z) f-+ fz(t) of (M 0 N) x Pinto M 0 N I8i P. This is bilinear in t and z and therefore induces a homomorphism

I: (M I8i N) I8i P --+ M I8i N 0 P

such that/«x I8i y) ® z) = x 0 y I8i z. To construct g, consider the mapping (x, y, z) f-+ (x I8i y) 0 z of M x N

x Pinto (M I8i N) 0 P. This is linear in each variable and therefore induces a homomorphism

g: M I8i N I8i P ~ (M 0 N) 0 P

such that g(x I8i y I8i z) = (x I8i y) I8i z. Clearly log and g 0 I are identity maps, hence f and g are isomorphisms. _

Exercise 2.15. Let A, B be rings, let M be an.A-module, P a B-module and N an (A, B)-bimodule (that is, N is simultaneously an A-module and a B-module and the two structures are compatible in the sense that a(xb) = (ax)b lor all a E A, bE B, x E N). Then M I8iA N is naturally a B-module, N I8iB P an A-module, and we hal'e

Let f: M ~ M', g: N ~ N' be homomorphisms of A-modules. Define h: M x N ~ M' I8i N' by h(x, y) = I(x) I8i g(y). It is easily checked that h is A-bilinear and therefore induces an A-module homomorphism

f I8i g: M I8i N ~ At' 0 N'

such that

(f I8i g)(x I8i y) = I(x) ® g(y) (XE M, YEN).

Let /': M' ~ M" and g': N' ~ N" be homomorphisms of A-modules. Then clearly the homomorphisms (f' 0 f) I8i (g' 0 g) and (f' I8i g') 0 (f ® g) agree on all elements of the form x I8i y in M I8i N. Since these elements generate M ® N, it follows that

(I' 0 f) 0 (g' 0 g) = (/' ® g') 0 (f 0 g).

RESTRICTION AND EXTENSION OF SCALARS

Let/: A ~ B be a homomorphism of rings and let N be a B-module. Then N has an A-module structure defined as follows: if a E A and x E N, then ax is de­fined to be I(a)x. This A-module is said to be obtained from N by restriction 01 scalars. In particular, I defines in this wayan A-module structure on B.

28 MODULES

Proposition 2.16. Suppose N isjinitely generated as a B-module and that B is finitely generated as an A-module. Then N isfinitely generated as an A-module.

Proof Let Yl, ... , Yn generate N over B, and let Xl, .•. , Xm generate B as an A-module. Then the mn products XiYJ generate N over A. _

Now let M be an A-module. Since, as we have just seen, B can be regarded as an A-module, we can form the A-module MB = B I8IA M. In fact MB carries a B-module structure such that b(b' 181 x) = bb' 181 X for all b, b' E B and all x E M. The B-module MB is said to be obtained from M by extension of scalars.

Proposition 2.17. If M is finitely generated as an A-module, then MB is finitely generated as a B-module.

Proof If Xl, ... , Xm generate M over A, then the I 0 Xi generate MB over B. _

EXACTNESS PROPERTIES OF THE TENSOR PRODUCT

Letf: M x N -+ P be an A-bilinear mapping. For each XE M the mapping y f-+ f(x, y) of N into P is A-linear, hence f gives rise to a mapping M-+ Hom (N, P) which is A-linear because f is linear in the variable x. Conversely any A-homomorphism cP: M -+ HomA (N, P) defines a bilinear map, namely (x, y) f-+ cp(x)(y). Hence the set S of A-bilinear mappings M x N -+ P is in natural one-to-one correspondence with Hom (M, Hom (N, P»). On the other hand S is in one-to-one correspondence with Hom (M 181 N, P), by the de­fining property of the tensor product. Hence we have a canonical isomorphism

Hom (M 181 N, P) ~ Hom (M, Hom (N, P»). (1)

Proposition 2.18. Let

M'~ M4 M"-+O (2)

be an exact sequence of A-modules and homomorphisms, and let N be any A-module. Then the sequence

M' 0 N~M 0N~ Mil I8IN-+O (3)

(where 1 denotes the identity mapping on N) is exact.

Proof Let E denote the sequence (2), and let E 0 N denote the sequence (3). Let P be any A-module. Since (2) is exact, the sequence Hom (E, Hom (N, P») is exact by (2.9); hence by (1) the sequence Hom (E 181 N, P) is exact. By (2.9) again, it follows that E 181 N is exact. _

Remarks. i) Let T(M) = M 0 N and let U(P) = Hom (N, P). Then (1) takes the form Hom (T(M), p) = Hom (M, U(P») for all A-modules M and P. In the language of abstract nonsense, the functor Tis the left adjoint of U, and U is the right adjoint of T. The proof of (2.18) shows that any functor which is a left adjoint is right exact. Likewise any functor which is a right adjoint is left exact.

ALGEBRAS 29

ii) It is not in general true that, if M' ~ M ~ Mit is an exact sequence of A-modules and homomorphisms, the sequence M' I8i N ~ M I8i N ~ Mit 0 N obtained by tensoring with an arbitrary A-module N is exact.

Example. Take A = Z and consider the exact sequence 0 -+ Z ~ Z, where ":'(x) = 2x for all x E Z. If we tensor with N = Z/2Z, the sequence 0 ~ Z 0 IIj

~ Z I8i N is not exact, because for any x I8i Y E Z I8i N we have

(f I8i l)(x I8i y) = 2x 0 y = x 0 2y = x I8i 0 = 0,

so that f 0 1 is the zero mapping, whereas Z I8i N # O.

The functor TN:M 1-+ M 0A N on the category of A-modules and homo­morphisms is therefore not in general exact. If TN is exact, that is to say if tensoring with N transforms all exact sequences into exact sequences, then N is said to be a fiat A-module.

Proposition 2.19. The following are equivalent, for an A-module N:

i) N isfiat.

ii) If 0 -+ M' -+ M -+ Mil -+ 0 is any exact sequence of A-modules, the tensored sequence 0 -+ M' 0 N -+ M 0 N ~ Mil I8i N -+ 0 is exact.

iii) Iff: M' -+ M is injective, then f I8i 1: M' I8i N -+ M I8i N is injective.

iv) Iff: M' -+ M is injective and M, M' are finitely generated, then f I8i 1: M' I8i N -+ M I8i N is injective.

Proof i) ~ ii) by splitting up a long exact sequence into short exact sequences.

ii) ~ iii) by (2.18).

iii).=> iv): clear.

iv) => iii). Letf: M' -+ M be injective and let u = 2: Xj I8i Yi E Ker (f I8i 1), so that 2:f(xD 0)'t = 0 in M I8i N. Let M~ be the submodule of M' generated by the x; and let Uo denote 2: x; 0 Yj as an element of M~ I8i N. By (2.14) there exists a finitely generated submodule Mo of M containingf(Mb) and such that 2:f(x;) 0 Yj = 0 as an element of Mo I8i N. If fo: Mb ~ Mo is the restriction off, this means that (fo 0 l)(uo) = O. Since Mo and M~ are finitely generated, fo I8i 1 is injective and therefore Uo = 0, hence u = O. •

Exercise 2.20. Iff" A -+ B is a ring homomorphism and Mis afiat A-module, then MB = E I8iA Mis afiat B-module. (Use the canonical isomorphisms (2.14), (2.15).)

ALGEBRAS

Letf: A -+ B be a ring homomorphism. If a E A and bEE, define a product

ab = f(a)b.

30 MODULES

This definition of scalar multiplication makes the ring B into an A-module (it is a particular example of restriction of scalars). Thus B has an A-module structure as well as a ring structure, and these two structures are compatible in a sense which the reader will be able to formulate for himself. The ring B, equipped with this A-module structure, is said to be an A-algebra. Thus an A-algebra is, by definition, a ring B together with a ring homomorphism f: A --+ B.

Remarks. i) In particular, if A is a field K (and B # 0) then f is injective by (1.2) and therefore K can be canonically identified with its image in B. Thus a K-algebra (K a field) is effectively a ring containing K as a subring.

ii) Let A be any ring. Since A has an identity element there is a unique homomorphism of the ring of integers Z into A, namely n 1--+ n.l. Thus every ring is automatically a Z-algebra.

Let f: A --+ B, g: A --+ C be two ring homomorphisms. An A-algebra homomorphism h: B --+ C is a ring homomorphism which is also an A-module homomorphism. The reader should verify that h is an A-algebra homomor­phism if and only if h 0 f = g.

A ring homomorphism f: A --+ B is finite, and B is a finite A-algebra, if B is finitely generated as an A-module. The homomorphismfis offinite type, and B is afinitely-generated A-algebra, ifthere exists a finite set of elements Xl> ..• Xn

in B such that every element of B can be written as a polynomial in Xl, ... , Xn

with coefficients inf(A); or equivalently if there is an A-algebra homomorphism from a polynomial ring A [t1> ... , tn ] onto B.

A ring A is said to be finitely generated if it is finitely generated as a Z­algebra. This means that there exist finitely many elements Xl> ... , X" in A such that every element of A can be written as a polynomial in the Xj with rational integer coefficients.

TENSOR PRODUcr OF ALGEBRAS

Let B, C be two A-algebras, f: A --+ B, g: A --+ C the corresponding homo­morphisms. Since Band C are A-modules we may form their tensor product D = B ®A C, which is an A-module. We shall noW define a multiplication on D.

Consider the mapping B x C x B x C --+ D defined by

(b, e, b', e') 1-+ bb' I8i ee'.

This is A-linear in each factor and therefore, by (2.12*), induces an A-module homomorphism

B ® C I8i B I8i C --+ D,

hence by (2.14) an A-module homomorphism

D®D--+D

EXERCISES 31

and this in turn by (2.11) corresponds to an A-bilinear mapping

p,: D x D~ D

which is such that

p,(b ® c, b' I8i c') = bb' ® cc'.

Of course, we could have written down this formula directly, but without some such argument as we have given there would be no guarantee that p, was well­defined.

We have therefore defined a multiplication on the tensor product D == B ®A C: for elements of the form b l8i'c it is given by

(b ® c)(b' ® c') = bb' ® cc',

and in general by

( L: (bi I8i ci») ( L: (bj I8i c;»)' = L: (bib; I8i CiC;). i 1 i,l

The reader should check that with this multiplication D is a commutative ring, with identity element 1 ® 1. Furthermore, D is an A-algebra: the mapping a 1-+ f(a) ® g(a) is a ring homomorphism A ~ D.

In fact there is a commutative diagram of ring homomorphisms

B

y "'",,, A/ ~D '" /' g~ /v

C

in which u, for example, is defined by u(b) = b ® 1.

EXERCISES

1. Show that (Z/mZ) @z(Z/nZ) = 0 if m, n are coprime.

2. Let A be a ring, a an ideal, M an A-module. Show that (A/a) @A M is isomor­phic to M/aM. [Tensor the exact sequence 0 -+ a -+ A ->- A/a -->- 0 with M,]

3. Let A be a local ring, M and N finitely generated A-modules. Prove that if M @ N = 0, then M == 0 or N == O. [Let m be the maximal ideal, k = A/m the residue field. Let Mk = k @A M ;;;; M/mM by Exercise 2. By Nakayama's lemma, Mk = 0 => M = O. But M @A N = 0 => (M @A N)k == 0 => M k @k Nk = 0 => Mk = 0 or Nk = 0, since M k , Nk are vector spaces over a field.]

4. Let M, (i E I) be any family of A-modules, and let M be their direct sum. Prove that M is flat -= each M, is flat.

2·

32 MODULE3

5. Let A[x] be the ring of polynomials in one indeterminate over a ring A. Prove that A[x] is a flat A-algebra. [Use Exercise 4.]

6. For any A-module, let M[x] denote the set of all polynomials in x with co­efficients in M. that is to say expressions of the form

Defining the product of an element of A [x] and an element of M[x] in the obvious way, show that M[x] is an A[x]-module.

Show that M[x] ~ A[x] @A M.

7. Let.p be a prime ideal in A. Show that p[x] is a prime ideal in A[x]. If m is a maximal ideal in A, is m[x] a maximal ideal in A[x]?

8. i) If M and N are flat A-modules, then so is M @A N. ii) If B is a flat A-algebra and N is a flat B-module, then N is flat as an A-module.

9. Let 0 -+ M' -+ M -+ M" -+ 0 be an exact sequence of A-modules. If M' and M" are finitely generated, then so is M.

10. Let A be a ring, a an ideal contained in the Jacobson radical of A; let M be an A-module and N a finitely generated A-module, and let u: M -+ N be a homo­morphism. If the induced homomorphism MlaM -+ NlaN is surjective, then u is surjective.

11. Let A be a ring i:- O. Show that Am ~ An => m = n. [Let m be a maximal ideal of A and let r/>: Am -+ An be an isomorphism. Then 1 @ r/>: (Aim) @ Am ---+ (Aim) @ An is an isomorphism between vector spaces of dimensions m and n over the field k = Aim. Hence m = n.] (Cf. Chapter 3, Exercise 15.)

If </>: Am -+ An is surjective, then m ;a. n. If r/>: Am -+ An is injective, is it always the case that m ~ n?

12. Let M be a finitely generated A-module and r/>: M -+ An a surjective homo­morphism. Show that Ker (r/» is finitely generated. [Let el,"" en De a basis of An and choose u, E M such that r/>(u,) = e, (1 ~ i ~ n). Show that M is the direct sum of Ker (r/» and the submodule generated by Ul,"" un.]

13. Let f: A ---+ B be a ring homomorphism, and let N be a B-module. Regarding N as an A-module by restriction of scalars, form the B-module NB = B @A N. Show that the homomorphism g: N -+ NB which maps y to 1 @ y is injective and that g(N) is a direct summand of N B •

[Definep: NB -+ Nby pCb @ y) = by, and show that NB = Im (g) EEl Ker (p).]

Direct limits

14. A partially ordered set I is said to be a directed set if for each pair i, j in I there exists k E I such that i ~ k and j ~ k.

Let A be a ring, let I be a directed set and let (M')'Ei be a family of A-modules indexed by·I. For each pair i,j in I such that i ~ j, let f.Ltf: M, ---+ Mf be an A-homomorphism, and suppose that the following axioms are satisfied:

(1) fJ-1I is the identity mapping of Mh for all i E I; (2) fJ-lk = fJ-jk 0 fJ-lj whenever i ~ j ~ k.

EXERCISES 33

Then the modules MI and homomorphisms I'-If are said to form a direct system M = (Mt, fJ-1f) over the directed set I.

We shall construct an A-module M called the direct limit of the direct system M. Let C be the direct sum of the Mt, and identify each module MI with its canonical image in C. Let D be the submodule of C generated by all elements of the form Xt - I'-If(Xt) where i ~ j and Xt E M t. Let M = Cf D, let fJ-: C -+ M

be the projection and let I'-t be the restriction of I'- to M t.

The module M, or more correctly the pair consisting of M and the family of homomorphisms f.LI: M t ~ M, is called the direct limit of the direct system M, and is written lim M t• From th~ construction it is clear that fJ-t = fJ-j 0 fJ-Ij

--;. whenever i ~ j.

15. In the situation of Exercise 14, show that every element of M can be written in the form fJ-,(xt) for some i E I and some XI E M t•

Show also that if fJ-tCXt) = 0 then there exists j ~ i such that fJ-1f(XI) = 0 in M j •

16. Show that the direct limit is characterized (up to isomorphism) by the following property. Let N be an A-module and for each i E I let at: M, ~ N be an A­module homomorphism such that a, = aj 0 I'-If whenever i ~ j. Then there exists a unique homomorphism a: M -+ N such that at = a 0 I'-t for all i E I.

17. Let (Mt)tel be a family of submodules of an A-module, such that for each pair of indices i, j in I there exists k E I such that M t + M j S M k • Define i ~ j to mean M, S M j and let fJ-1f: M -+ M j be the embedding of M t in M j • Show that

lim M t = 2: M = U M,. ---+

In particular, any A-module is the direct limit of its finitely generated sub­modules.

18. Let M = (Mt, I'-tf), N = (Nt, Vtj) be direct systems of A-modules over the same directed set. Let M, N be the direct limits and I'-t: Mt -+ M, V,: Nt -+ N the associated homomorphisms.

A homomorphism <1>: M -+ N is by definition a family of A-module homo­morphisms </>t: M t -+ Nt such that cPj 0 fJ-tj = V'f 0 cP, whenever i ~ j. Show that <I> defines a unique homomorphism </> = lim </>,: M -+ N such that </> 0 I'-t =

---+ Vt 0 </>t for all i E I.

19. A sequence of direct systems and homomorphisms

M-+N-+P

is exact if the corresponding sequence of modules and module homomorphisms

is exact for each i E I. Show that the sequence M N·--+ P of direct limits is then exact. [Use Exercise 15.]

Tensor products commute with direct limits

20. Keeping the same notation as in Exercise 14, let N be any A-module. Then (M 181 N,I'-If 181 1) is a direct system; let P = lim (Mt 181 N) be its direct limit.

---i>-

34 MODULES

For each i E Iwe have a homomorphism /J-I 0 1: M j 0 N -+ M 0 N, hence by Exercise 16 a homomorphism .p: P ---+ M 0 N. Show that .p is an isomorphism, so that

lim (Mj 0 N) ~ (lim M j ) 0 N. ~ ~

[For each i E I, let gj: M j x N -+- M j 0 N be the canonical bilinear mapping. Passing to the limit we obtain a mapping g: M x N -+ P. Show that g is A-bilinear and hence define a homomorphism r/>: M 0 N ---+ P. Verify that r/> 0 .p and .p 0 r/> are identity mappings.]

21. Let (Aj)jel be a family of rings indexed by a directed set I, and for each pair i ~ j in I let ajl: A j -+ A, be a ring homomorphism, satisfying conditions (1) and (2) of Exercise 14. Regarding each A j as a Z-module we can then form the direct limit A = lim A j. Show that A inherits a ring structure from the A j so that the

~

mappings AI ---+ A are ring homomorphisms. The ring A is the direct limit of the system (Ab CZjI)'

If A = 0 prove that AI = 0 for some i E l. [Remember that all rings have identity elements!]

22. Let (Ah all) be a direct system of rings and let 91 j be the nilradical of A j • Show that lim 91j is the nilradical of lim A j •

~ ~

If each AI is an integral domain, then lim AI is an integral domain. ~

23. Let (B"heA be a family of A-algebras. For each finite subset of A let BJ denote the tensor product (over A) of the B" for ,\ E J. If J' is another finite subset of A and J S J', there is a canonical A-algebra homomorphism B,-+ B". Let B denote the direct limit of the rings B, as J runs through all finite subsets of A. The ring B has a natural A-algebra structure for which the homomorphisms B, -+ B are A-algebra homomorphisms. The A-algebra B is the tensor product of the family (B"')"eA'

Flatness and Tor In these Exercises it will be assumed that the reader is familiar with the definition and basic properties of the Tor functor.

24. If M is an A-module, the following are equivalent: i) M is flat;

ii) Tor~ (M, N) = 0 for all n > 0 and all A-modules N; iii) Tort (M, N) = 0 for all A-modules N. [To show that (i) => (ii), take a free resolution of N and tensor it with M. Since M is flat, the resulting sequence is exact and therefore its homology groups, which are the Tor~ (M, N), are zero for n > O. To show that (iii) => (i), let 0-+ N' -+ N -+ N" -+ 0 be an exact sequence. Then, from the Tor exact sequence,

TorI (M, Nj ---+ M 0 N' ---+ M 0 N -+ M 0 N" -+ 0

is exact. Since TorI (M, Nj = 0 it follows that M is flat.]

25. Let 0 -+ N' -+ N -+ N" -+ 0 be an exact sequence, with N" flat. Then N' is flat -¢> N is flat. [Use Exercise 24 and the Tor exact sequence.]

EXERCISES 35

26. Let N be an A-module. Then N is flat -=> TorI (A/a, N) = 0 for all finitely generated ideals a in A. [Show first that N is flat if Torl (M, N) = 0 for all finitely generated A-modules M, by using (2.19). If M is finitely generated, let Xl, ... , X,. be a set of generators of M, and let M, be the submodule generated by Xl, ... , X,. By considering the successive quotients Md M, -1 and using Exercise 25, deduce that N is fiat if Torl (M, N) = 0 for all cyclic A-modules M, i.e., all M generated by a single element, and therefore of the form A/a for some ideal a. Finally use (2.19) again to reduce to the case where a is a finitely generated ideal.]

27. A ring A is absolutely flat if every A-module is flat. Prove that the following are equivalent: i) A is absolutely flat.

ii) Every principal ideal is idempotent. iii) Every finitely generated ideal is a direct summand of A. [i) => ii). Let X E A. Then A/(x) is a flat A-module, hence in the diagram

(x) @ A ~ (x) @ A/(x) ,j, ,j,Q

A -- A/(x)

the mapping a is injective. Hence 1m (13) = 0, hence (x) = (X2). ii) => iii). Let x E A. Then x = ax2 for some a E A, hence e = ax is idempotent and we have (e) = (x). Now if e.!are idempotents, then (e.!) = (e + f - e/). Hence every finitely generated ideal is principal, and generated by an idempotent e, hence is a direct summand because A = (e) EB (1 - e). iii) => i). Use the criterion of Exercise 26.]

28. A Boolean ring is absolutely flat. The ring of Chapter I, Exercise 7 is absolutely flat. Every homomorphic image of an absolutely fiat ring is absolutely fiat. If a local ring is absolutely flat, then it is a field.

If A is absolutely flat, every non-unit in A is a zero-divisor.

3

Rings and Modules of Fractions

The formation of rings of fractions and the associated process of localization are perhaps the most important technical tools in commutative algebra. They correspond in the algebro-geometric picture to concentrating attention on an open set or near a point, and the importance of these notions should be self­evident. This chapter gives the definitions and simple properties of the formation of fractions.

The procedure by which one constructs the rational field Q from the ring of integers Z (and embeds Z in Q) extends easily to any integral domain A and produces the field of fractions of A. The construction consists in taking all ordered pairs (a, s) where a, SEA and s i= 0, and setting up an equivalence relation between such pairs:

(a,s) == (b,t) <=>at - bs = O.

This works only if A is an integral domain, because the verification that the relation is transitive involves canceling, i.e. the fact that A has no zero-divisor i= o. However, it can be generalized as follows:

Let A be any ring. A multiplicatively closed subset of A is a subset S of A such that 1 E Sand S is closed under multiplication: in other words S is a sub­semigroup of the multiplicative semigroup of A. Define a relation == on A x S as follows:

(a, s) == (b, t) <=> (at - bs)u = 0 for some U E S.

Clearly this relation is reflexive and symmetric. To show that it is transitive, suppose (a, s) == (b, t) and (b, t) == (c, u). Then there exist v, w in S such that (at - bs)v = 0 and (bu - ct)w = o. Eliminate b from these two equations and we have (au - cs)tvw = O. Since S is closed Under multiplication, we have tvw E S, hence (a, s) == (c, u). Thus we have an equivalence relation. Let a/s denote the equivalence class of (a, s), and let S -1 A denote the set of equivalence classes. We put a ring structure on S -lA by defining addition and multiplication of these "fractions" a/s in the same way as in elementary algebra: that is,

(a/s) + (b/t) = (at + bs)/st, (a/s)(b/t) = ab/st.

36

RINGS AND MODULES OF FRACTIONS 37

Exercise. Verify that these definitions are independent of the choices of rep­resentatives (a, s) and (b, t), and that S -1 A satisfies the axioms of a com­mutative ring with identity.

We also have a ring homomorphism f: A -+ S -1 A defined by f(x) = x/i. This is not in general injective.

Remark. If A is an integral domain and S = A - {O}, then S -1 A is the field of fractions of A.

The ring S -1 A is called the ring of fractions of A with respect to S. It has a universal property:

Proposition 3.1. Let g: A -+ B be a ring homomorphism such that g(s) is a unit in B for all s E S. Then there exists a unique ring homomorphism h: S -1 A -+ B such that g = h 0 f

Proof i) Uniqueness. If h satisfies the conditions, then heal!) = hj(a) = g(a) for all a E A; hence, if s E S,

h(l/s) = h(s/l)-l) = h(s/l)-l = g(S)-l

and therefore heals) = h(a/l)·h(1/s) = g(a)g(s)-I, so that h is uniquely determined by g.

ii) Existence. Let heals) = g(a)g(s)-l. Then h will clearly be a ring homo­morphism provided that it is well-defined. Suppose then that a/s = a'/s'; then there exists t E S such that (as' - a's)t = 0, hence

(g(a)g(s') - g(a')g(s»g(t) = 0;

now get) is a unit in B, hence g(a)g(s)-l = g(a')g(s')-l. •

The ring S-lA and the homomorphismf: A -+ S-lA have the following properties:

1) s E S => f(s) is a unit in S-lA;

2) f(a) = 0 => as = 0 for some s E S;

3) Every element of S -1 A is of the formf(a)f(s) -1 for some a E A and some SES.

Conversely, these three conditions determine the ring S -1 A up to iso­morphism. Precisely:

Corollary 3.2. Ifg: A -+ B is a ring homomorphism such that

i) s E S => g(s) is a unit in B;

ii) g(a) = 0 => as = o for some SE S;

38 RINGS AND MODULES OF FRACTIONS

iii) Every element of B is of the form g(a)g(s) -1 ; then there is a unique isomorphism h: 8- 1A ~ B such that g = h 0 f

Proof By (3.1) we have to show that h: 8 -1 A ~ B, defined by

h(als) = g(a)g(s)-l

(this definition uses i») is an isomorphism. By iii), h is surjective. To show h is injective, look at the kernel of h: if h(als) = 0, then g(a) = 0, hence by ii) we have at = ° for some t E 8, hence (a, s) == (0, 1), i.e., als = ° in 8- 1A. •

Examples. 1) Let p be a prime ideal of A. Then 8 = A - .p is multiplicatively closed (in fact A - P is multiplicatively closed ~ .p is prime). We write All for 8 -lAin this case. The elements als with a E.p form an ideal m in All' If bit ~ m, then b ~.p, hence b E 8 and therefore bit is a unit in All' It follows that if a is an ideal in All and a $ m, then a contains a unit and is therefore the whole ring. Hence m is the only maximal ideal in All; in other words, All is a local ring.

The process of passing from A to All is called localization at .p.

2) 8- 1A is the zero ring ~ ° E 8. 3) LetfE A and let 8 = {rn},,~o. We write A, for 8- 1A in this case. 4) Let a be any ideal in A, and let 8 = 1 + a = set of all 1 + x where

x E a. Clearly 8 is multiplicatively closed. 5) Special cases of 1) and 3): i) A = Z, P = (P), p a prime number; All = set of all rational numbers

min where n is prime to p; iffE Z andf:F 0, then A, is the set of all rational numbers whose denominator is a power off

ii) A = k[tl>' .. , t,,], where k is a field and the t( are independent indeter­minates, p a prime ideal in A. Then All is the ring of all rational functionsJ7g, where g ~.p. If V is the variety defined by the ideal .p, that is to say the set of all x = (Xl" .. , x,,) E k" such that f(x) = 0 whenever fE.p, then (provided k is infinite) All can be identified with the ring of all rational functions on k" which are defined at almost all points of V; it is the local ring of k" along the variety V. This is the prototype of the local rings which arise in algebraic geometry.

The construction of 8- 1A can be carried through with an A-module Min place of the ring A. Define a relation == on M x 8 as follows:

(m, s) == (m', s') ~ 3t E 8 such that t(sm' - s'm) = O.

As before, this is an equivalence relation. Let m/s denote the equiValence class of the pair (m, s), let 8 -1 M denote the set of such fractions, and make 8 -1 M into an 8-1A-module with the obvious definitions of addition and scalar multiplication. As in Examples 1) and 3) above, we write Mil instead of S-IM when 8 = A - .p (.p prime) and M, when 8 = {f"},,>o.

Let u: M ~ N be an A-module homomorphism. Then it gives rise to an 8- 1A-module homomorphism 8- 1u: 8- 1M ~ S-lN, namely 8- 1u maps mls to u(m)/s. We have 8 -lev 0 u) = (8 -lV) 0 (8 -lU).

RINGS AND MODULES OF FRACTIONS 39

Proposition 3.3. The operation S-l is exact, i.e., if M' ~ M ~ M- is exact at M, then S-lM' S-1/) S-lM S-1,,) S-lM- is exact at S-lM,

Proof Wehavegof= 0,henceS-1goS-1f= S-l(O) = O,hencelm(S-1j) s; Ker (S -lg). To prove the reverse inclusion, let m/s E Ker (S -lg), then g(m)/s = 0 in S -1 M H

, hence there exists t E S such that tg(m) = ° in M-, But tg(m) = g(tm) since g is an A-module homomorphism, hence tm E Ker (g) = 1m (f) and therefore tm = f(m') for some m' EM', Hence in S-IM we have m/s = f(m')/st = (S-1J)(m'/st) Elm (S-l!). Hence Ker (S-lg) s; 1m (S-1j),

• In particular, it follows from (3.3) that if M' is a submodule of M, the map­

ping S-lM' ~ S-lM is injective and therefore S-lM' can be regarded as a submodule of S-IM. With this convention,

Corollary 3.4. Formation of fractions commutes with formation of finite sums, finite intersections and quotients, Precisely, if N, Pare submodules of an A-module M, then

i) S-l(N + P) = S-l(N) + S-l(P)

ii) S-l(N n P) = S-l(N) n S-l(P)

iii) the S-lA-modules S-l(M/N) and (S-IM)/(S-lN) are isomorphic,

Proof i) follows readily from the definitions and ii) is easy to verify: if y/s = zit (y E N, z EP, s, t E S) then u(ty - sz) = ° for some u E S, hence w = uty = USZ EN n P and therefore y/s = w/stu E S -l(N n P). Consequently S-l N n S -lp s; S -leN n P), and the reverse inclusion is obvious.

iii) Apply S -1 to the exact sequence ° ~ N ~ M ~ M/N ~ 0. •

Proposition 3.5. Let M be an A -module. Then the S -1 A modules S -1M and S-lA 0A M are isomorphic; more precisely, there exists a unique iso­morphismf: S-IA 0A M ~ S-lMfor which

f(a/s) 0 m) = am/s for all a E A, m E M, s E S. (1)

Proof The mapping S-lA x M ~ S-lM defined by

(a/s, m) ~ am/s

is A-bilinear, and therefore by the universal property (2.12) of the tensor product induces an A-homomorphism

f: S-lA 0A M ~ S-lM

satisfying (1). Clearly fis surjective, and is uniquely defined by (1). Let 2:1 (adsl) 0 ml be any element of S -1 A 0 M. If s = TIl Sl E S,

t l = TIf"l Sf' we have

40 RINGS AND MODULES OF FRACTIONS

so that every element of S-lA 0 Mis of the form (l/s) 0 m. Suppose that f((I/s) 0 m) = O. Then m/s = 0, hence tm = 0 for some t E S, and therefore

I tIl -0m=-0m=-0~=-00=Q s st st st

Hence f is injective and therefore an isomorphism. _

Corollary 3.6. S -1 A is a flat A-module. Proof (3.3), (3.5). _

Proposition 3.7. If M, N are A-modules, there is a unique isomorphism of S-lA-modulesf: S-lM 0S- 1 A S-lN ~ S-l(M 0A N)such that

f(m/s) 0 (n/t») = (m 0 n)/st.

In particular, if p is any prime ideal, then

MtJ 0AtJ NtJ ~ (M 0A N)tJ

as AtJ-modules.

Proof Use (3.5) and the canonical isomorphisms of Chapter 2. _

LOCAL PROPERTIES

A property P of a ring A (or of an A-module M) is said to be a local property if the following is true:

A (or M) has P <:> AtJ (or MtJ) has P, for each prime ideal p of A. The following propositions give examples of local properties:

Proposition 3.8. Let M be an A-module. Then the following are equivalent:

i) M = 0;

ii) M tJ = 0 for all prime ideals p of A;

iii) Mm = 0 for all maximal ideals m of A.

Proof Clearly i) ~ ii) Q iii). Suppose iii) satisfied and M :F O. Let x be a non-zero element of M, and let a = Ann (x); a is an ideal :F (1), hence is contained in a maximal ideal m by (1.4). Consider x/I E Mm. Since Mm = 0 we have x/I = 0, hence x is killed by some element of A - m; but this is impossible since Ann (x) s; m. _

Proposition 3.9. Let 1>: M ~ N be an A-module homomorphism. following are equivalent:

• i) 1> is injective;

ii) 1>tJ: MtJ ~ NtJ is injective for each prime ideal p;

iii) 1>m: Mm ~ Nut is injective for each maximal ideal m.

Simiklrly with "injective" replaced by "surjective" thr~ughout.

Then the

EXTENDED AND CONTRACTED IDEALS IN RINGS OF FRACTIONS 41

Proof i) ~ ii). 0 ~ M ~ N is exact, hence 0 ~ MtJ ~ NtJ is exact, i.e., 1>tJ is injective.

ii) ~ iii) because a maximal ideal is prime.

iii) ~ i). Let M' = Ker (1)), then the sequence 0 ~ M' ~ M ~ N is exact, hence 0 ~ M'm ~ Mm ~ Nm is exact by (3.3) and therefore M'm ~ Ker (1)m) = 0 since 1>m is injective. Hence M' = 0 by (3.8), hence 1> is injective.

F or the other part of the proposition, just reverse all the arrows. _

Flatness is a local property:

Proposition 3.10. For any A-module M, the following statements are equivalent:

i) Mis aflat A-module:

ii) M tJ is a flat AtJ-module for each prime ideal p;

iii) M m is a flat Am-module for each maximal ideal m.

Proof i) ~ ii) by (3.5) and (2.20).

ii) ~ iii) O.K.

iii) ~ i). If N ~ P is a homomorphism of A-modules, and m is any maximal ideal of A, then

N ~ P injective ~ Nm ~ Pm injective, by (3.9) ~ Nm Q9Am Mm ~ Pm Q9Am Mminjective, by (2.19)

~ (N Q9A M)m ~ (P Q9A M)m injective, by (3.7) ~ N Q9A M ~ P Q9A M injective, by (3.9).

Hence M is flat by (2.19). _

EXTENDED AND CONTRACTED IDEALS IN RINGS OF FRACTIONS

Let A be a ring, S a multiplicatively closed subset of A and f: A ~ S -lA the natural hom~morphism, defined by f(a) = a/I. Let C be the set of contracted ideals in A, and let E be the set of extended ideals in S -1 A (cf. (1.17»). If a is an ideal in A, its extension ae in S -1 A is S -la (for any y E ae is of the form Lads!> where aj E a and Sj E S; bring this fraction to a common denominator).

Proposition 3.11. i) Every ideal in S -1 A is an extended ideal. '

ii) If a is an ideal in A, then aee = Uses (a:s). Hence a" = (1) if and only if a meets S.

iii) a E C <:> no element of S is a zero-divisor in A/a.

iv) The prime ideals of S -1 A are in one-to-one correspondence (p +--+ S -lp) with the prime ideals of A which don't meet S.

42 RINGS AND MODULES OF FRACTIONS

v) The operation S - 1 commutes with formation of finite sums, products, intersections and radicals.

Proof i) Let b be an ideal in S-lA, and let x/s E b. Then x/I E b, hence x E bC and therefore x/s E bee. Since b ;;2 bce in any case (1.17), it follows that b = bce.

ii) x E aec = (S-la)C <:> x/I = a/s for some a E a, SE S <:> (xs - a)t = 0 for some t E S <:> xst E a <:> X E UseS (a:s).

iii) a E C <:> aec S; a <:> (sx E a for some s E S ~ X E a) <:> no s E S is a zero-divisor in A/a.

iv) If q is a prime ideal in S-lA, then qC is a prime ideal in A (this much is true for any ring homomorphism). Conversely, if t> is a prime ideal in A, then A/t> is an integral domain; if S is the image of S in Aft>, we have S -IA/S -It> ~ S -l(A/t» which is either 0 or else is contained in the field of fractions of A/t> and is therefore an integral domain, and therefore S -It> is either prime or is the unit ideal; by i) the latter possibility occurs if and only if t> meets S.

v) For sums and products, this follows from (1.18); for intersections, from (3.4). As to radicals, We have S-lr(a) S;; r(S-la) from (1.18), and the proof of the reverse inclusion is a routine verification which we leave to the reader. _

Remarks. 1) If a, b are ideals of A, the formula

S-l(a:b) = (S-la:S-lb)

is true provided the ideal b is finitely generated: see (3.15).

2) The proof in (1.8) that if fE A is not nilpotent there is a prime ideal of A which does not contain f can be expressed more concisely in the language of rings of fractions. Since the set S = (j"),,~o does not contain 0, the ring S-IA = A, is not the zero ring and therefore by (1.3) has a maximal ideal, whose contraction in A is a prime ideal t> which does not meet S by (3.11); hencef~ t>.

Corollary 3.12. If'iR is the nilradical of A, the nilradical of S -1 A is S -19C. •

Corollary 3.13. If t> is a prime ideal of A, the prime ideals of the local ring Ap are in one-to-one correspondence with the prime ideals of A contained in t>.

Proof Take S = A - t> in (3.1 I) (iv). •

Remark. Thus the passage from A to Ap cuts out all prime ideals except those contained in t>. In the other direction, the passage from A to Aft> cuts out all prime ideals except those containing t>. Hence if t>, q are prime ideals such that t> ;;2 q, then by' localizing with respect to t> and taking the quotient mod q (in either order: these two operations commute, by (3.4»), we restrict Our atten­tion to those prime ideals which lie between t> and q. In particular, if t> = q we

EXERCISES 43

end up with a field, called the residue field at .\;), which can be obtained either as the field of fractions of the integral domain A/.\;) or as the residue field of the local ring All.

Proposition 3.14. Let M be a finitely generated A-module, S a multiplicatively closed subset of A. Then S-l(Ann (M») = Ann (S-lM).

Proof If this is true for two A-modules, M, N, it is true for M + N:

S-l(Ann(M + N») = S-l(Ann(M) (') Ann (N» by (2.2) = S-l(Ann (M») (') S-l(Ann (N») by (3.4) = Ann (S -1 M) (') Ann (S -IN) by hypothesis = Ann (S-lM + S-lN) = Ann (S-l(M + N»).

Hence it is enough to prove (3.14) for M generated by a single element: then M ~ A/a (as A-module), where a = Ann (M); S-lM ~ (S-lA)/(S-la) by (3.4), so that Ann (S-lM) = S-la = S-l(Ann (M») .•

Corollary 3.15. If N, Pare submodules of an A-module M and if P is finitely generated, then S -l(N: P) = (S -1 N: S -1 P).

Proof (N:P) = Ann (N + P)/N) by (2.2); now apply (3.14). •

Proposition 3.16. Let A ~ B be a ring homomorphism and let p be a prime ideal of A. Then p is the contraction of a prime ideal of B if and only if .\;)ec = .\;).

Proof If p = qC then pec = p by (1.17). Conversely, if pec = p, let S be the image of A - pin B. Then.\;)e does not meet S, therefore by (3.11) its extension in S -1 B is a proper ideal and hence is contained in a maximal ideal m of S -lB. If q is the contraction of m in B, then q is prime, q ;2 pe and q (') S = 0. Hence qC = p. •

EXEROSES

1. Let S be a multiplicatively closed subset of a ring A, and let M be a finitely generated A-module. Prove that S -1 M = 0 if and only if there exists S E S such that sM = o.

2. Let a be an ideal of a ring A, and let S = 1 + a. Show that S -la is contained in the Jacobson radical of S-IA.

Use this result and Nakayama's lemma to give a proof of (2.5) which does not depend on determinants. [If M = aM, then S-1M = (S-1a)(S-IM), hence by Nakayama we have S-IM = o. Now use Exercise 1.]

3. Let A be a ring, let Sand T be two multiplicatively closed subsets of A, and let U be the image of Tin S -1 A. Show that the rings (Sn -1 A and U -1(S -1 A) are

, isomorphic.

44 RINGS AND MODULES OF FRACTIONS

4. Let f: A -+ B be a homomorphism of rings and let S be a multiplicatively closed subset of A. Let T = f(S). Show that S -1 Band T-l B are isomorphic as S-lA-modules.

5. Let A be a ring. Suppose that, for each prime ideal .p, the local ring Ap has no nilpotent element i:- O. Show that A has no nilpotent element i:- O. If each A~ is an integral domain, is A necessarily an integral domain?

6. Let A be a ring i:- 0 and let ~ be the set of all multiplicatively closed subsets S of A such that 0 rI: S. Show that ~ has maximal elements, and that S E l: is maximal if and only if A - S is a minimal prime ideal of A.

7. A multiplicatively closed subset S of a ring A is said to be saturated if

xy E S ¢> XES and yES.

Prove that i) S is saturated ¢> A S is a union of prime ideals.

ii) If S is any multiplicatively closed subset of A, there is a unique smallest saturated multiplicatively closed subset S containing S, and that S is the complement in A of the union of the prime ideals which do not meet S. (S is called the saturation of S.)

If S = 1 + a, where a is an ideal of A, find S.

8. Let S, T be multiplicatively closed subsets of A, such that S S; T. Let c/>: S -1 A -+ T-IA be the homomorphism which maps each a/s E S-1A to a/s considered as an element of T-l A. Show that the following statements are equivalent: i) c/> is bijective.

ii) For each t E T, t/l is a unit in S ··1 A. iii) For each t E T there exists x E A such that xl E S. iv) T is contained in the saturation of S (Exercise 7). v) Every prime ideal which meets T also meets S.

9. The set So of all non-zero-divisors in A is a saturated multiplicatively closed subset of A. Hence the set D of zero-divisors in A is a union of prime ideals (see Chapter 1, Exercise 14). Show that every minimal prime ideal of A is contained in D. [Use Exercise 6.]

The ring So 1 A is called the total ring of fractions of A. Prove that i) So is the largest multiplicatively closed subset of A for which the homo­

morphism A -+ So 1 A is injective. * ii) Every element in So 1 A is either a zero-divisor or a unit.

iii) Every ring in which every non-unit is a zero-divisor is equal to its total ring of fractions (i.e., A -+ So 1 A is bijective).

10. Let A be a ring. i) If A is absolutely flat (Chapter 2, Exercise 27) and S is any multiplicatively

closed subset of A, then S-IA is absolutely Bat. ii) A is absolutely flat <:> Am is a field for each maximal ideal m.

11. Let A be a ring. Prove that the following are equivalent: i) A/~ is absolutely flat (91 being the nilradical of A).

ii) Every prime ideal of A is maximal.

EXERCISES 45

iii) Spec (A) is a T1-space (i.e., every subset consisting of a s'ngle point is closed). iv) Spec (A) is Hausdorff.

If these conditions are satisfied, show that Spec (A) is compact and totally disconnected (i.e. the only connected subsets of Spec (A) are those consisting of a single point).

11. Let A be an integral domain and M an A-module. An element x E M is a torsion element of M if Ann (x) -:j: 0, that is if x is killed by some non-zero element of A. Show that the torsion elements of M form a submodule of M. This submodule is called the torsion submodule of M and is denoted by T(M). If T(M) = 0, the module M is said to be torsion-free. Show that

i) If M is any A-module, then M/T(M) is torsion-free. ii) Iff: M -7 N is a module homomorphism, then f(T(M» S T(N).

iii) If 0 ->- M' -7 M -7 M" is an exact sequence, then the sequence 0 -7 T(M') -7 T(M) ->- T(M") is exact.

iv) If M is any A-module, then T(M) is the kernel of the mapping x I-t 1 181 x of Minto K I8IA M, where K is the field of fractions of A.

[For iv), show that K may be regarded as the direct limit of its submodules A g (g E K); using Chapter 1, Exercise 15 and Exercise 20, show that if 1 181 x = 0 in K 181 Mthen 1 181 x = 0 in Ag 181 Mfor some g #- O. Deduce that g-lx = '0.1

13. Let S be a multiplicatively closed subset of an integral domain A. In the notation of Exercise 12, show that T(S -1 M) = S -l(TM). Deduce that the following are equivalent: i) M is torsion-free. ii) M~ is torsion-free for all prime ideals p.

iii) Mm is torsion-free for all maximal ideals m.

14. Let M be an A-module and a an ideal of A. Suppose that Mm = 0 for all maximal ideals m ~ a. Prove that M = aM. [Pass to the A/a-module M/aM and use (3.8).J

15. Let A be a ring, and let F be the A-module An. Show that every set of n gen­erators of F is a basis of F. [Let Xl> ... , x" be a set of generators and el, ... , en the canonical basis of F. Define c/>: F -7 F by c/>(et) = Xt. Then c/> is surjective and we have to prove that it is an isomorphism. By (3.9) we may assume that A is a local ring. Let N be the kernel of c/> and let k = A /m be the residue field of A. Since F is a flat A-module, the exact sequence 0 ->- N -7 F ->- F -7 0 gives an

10<1> exactsequenceO~k 0 N~k I8IF~k I8IF~O. Nowk I8IF= k n

is an n-dimensional vector space over k; 1 181 c/> is surjective, hence bijective, hence k 0 N = O.

Also N is finitely generated, by Chapter 2, Exercise 12, hence N = 0 by Nakayama's lemma. Hence c/> is an isomorphism.]

Deduce that every set of generators of F has at least n elements.

16. Let B be a flat A-algebra. Then the following conditions are equivalent: i) aee = a for all ideals a of A.

ii) Spec (B) -+ Spec (A) is surjective. iii) For every maximal ideal m of A we have me -:j: (I).

46 RINGS AND MODULES OF FRACTIONS

iv) If M is any non-zero A-module, then MB #: O. v) For every A-module M, the mapping x f-+ 1 0 x of Minto MB is injective. [For i) => ii), use (3.16). ii) => iii) is clear. iii) => iv): Let x be a non-zero element of M and let M' = Ax. Since B is flat over A it is enough to show that Ms #: O. We have M' ;;;;; A/a for some ideal a #: (1), hence Ms ~ B/as• Now a s m for some maximal ideal m, hence a" S m" i: (1). Hence Ms #: O. iv) => v): Let M' be the kernel of M -+ M B • Since B is flat over A, the sequence 0-+ Ms -+ MB -+ (MB)B is exact. But (Chapter 2, Exercise 13, with N = M B) the mapping MB ~ (MBh is injective, hence Ms = 0 and therefore M' = O. v) => i): Take M = A/a.]

B is said to be faithfully flat over A.

17. Let A!.... B ~ C be ring homomorphisms. If go fis flat and g is faithfully flat, then f is flat.

18. Letf: A ~ B be a flat homomorphism of rings, let q be a prime ideal of B and let ~ = qC. Thenf*: Spec (Bq) -+ Spec (Ap) is surjective. [For Bp is flat over Ap by (3.10), and Bq is a local ring of Bp, hence is flat over Bp. Hence Bq is flat over Ap and satisfies condition (3) of Exercise 16.]

19. Let A be a ring, M an A-module. The support of M is defined to be the set Supp (M) of prime ideals .p of A such that Mp #: O. Prove the following results:

i) M #: 0 - Supp (M) #: 0. ii) V(a) = Supp (A/a).

iii) If 0 ~ M' ~ M ~ M" -+ 0 is an exact sequence, then Supp (M) = Supp (M') v Supp (M,,).

iv) If M = ~ M, then Supp (M) = U Supp (M1).

v) If M is finitely generated, then Supp (M) = V( Ann (M») (and is therefore a closed subset of Spec (A»).

vi) If M, N are finitely generated, then Supp (M 0A N) = Supp (M) Il Supp (N). - [Use Chapter 2, Exercise 3.]

vii) If M is finitely generated and a is an ideal of A, then Supp (M/aM) =

V(a + Ann (M»). viii) If f: A ~ B is a ring homomorphism and M is a finitely generated A­

module, then Supp (B 0A M) = f*-l(SUpp (M»).

20. Let f: A ~ B be a ring homomorphism, f*: Spec (B) -+ Spec (A) the associated mapping. Show that

i) Every prime ideal of A is a contracted ideal - f* is surjective. ii) Every prime ideal of B is an extended ideal => f* is injective.

Is the converse of ii) true?

21. i) Let A be a ring, S a multiplicatively closed subset of A, and c{>: A -+ S-lA the canonical homomorphism. Show that c{> *: Spec (S -1 A) -+ Spec (A) is a homeomorphism of Spec (S-lA) onto its image in X = Spec (A). Let this image be denoted by S -1 X. In particular, if fE A, the image of Spec (A,) in X is the basic open set X, (Chapter 1, Exercise 17).

EXERCISES 47

ii) Let f: A ->- B be a ring homomorphism. Let X = Spec (A) and Y = Spec (B), and let /*: Y -+ X be the mapping associated with f. Identifying Spec(S-lA) with its canonical image S-1X in X, and Spec(S-1B) (= Spec (J(S)-1B))withitscanonicalimageS-1Y in Y,show thatS-1/*: Spec (S-1 B) ->- Spec (S-1 A) is the restriction of f· to S-1 Y, and that S -1 Y = /*-1(S-1X).

iii) Let a be an ideal of A and let 0 = a" be its extension in B. Let/: A/a -+- BfO be the homomorphism induced by f If Spec (A/a) is identified with its canonical image V(a) in X, and Spec (BfO) with its image V(b) in Y, show that I· is the restriction off· to V(b).

iv) Let .p be a prime ideal of A. Take S = A - .p in ii) and then reduce mod S -l.p as in iii). Deduce that the subspace /* -1(p) of Y is naturally homeomorphic to Spec (Bp/.pBp) = Spec (k(p) 0A B), where k(.p) is the residue field of the local ring Ap. Spec (k(.p) 0A B) is called the fiber of /* over p.

22. Let A be a ring and p a prime ideal of A. Then the canonical image of Spec (Ap) in Spec (A) is equal to the intersection of all the open neighborhoods of .p in Spec (A).

23. Let A be a ring, let X = Spec (A) and let U be a basic open set in X (Le., U = XI for some f E A: Chapter 1, Exercise 17).

i) If U = X" show that the ring A(U) = AI depends only on U and not onf ii) Let U ' = Xg be another basic open set such that U ' s; U. Show that there

is an equation of the form gn = uffor some integer n > 0 and some U E A, and use this to define a homomorphism p: A(U) -+ A(U') (i.e., AI -+ Ag) by mapping a/f'" to au"'/gmn. Show that p depends only on U and U'. This homomorphism is called the restriction homomorphism.

iii) If U = U' , then p is the identity map. iv) If U ~ U ' ~ U" are basic open sets in X, show that the diagram

A(U) ) A(U,,) '>.. .l'

A(U')

(in which the arrows are restriction homomorphisms) is commutative. v) Let x (= .p) be a point of X. Show that

lim A(U) ~ All. ~ Ua"

The assignment of the ring A( U) to each basic open set U of X, and the restriction homomorphisms p, satisfying the conditions iii) and iv) above, constitutes a presheaf of rings on the basis of open sets (XI)/eA. v) says that the stalk of this presheaf at x E X is the corresponding local ring Ap.

24. Show that the presheaf of Exercise 23 has the following property. Let (UI)I&1 be a covering of X by basic open sets. For each i E I let Sl E A(UI) be such that, for each pair of indices i,j, the images of SI and Sf in A(UIIl Uf) are equal. Then there exists a unique SEA (= A(X») whose image in A(Ut) is Sh for all i E I. (This essentially implies that the presheaf is a sheaf)

48 RINGS AND MODULES OF FRACTIONS

25. Let I: A ->- B, g: A -+ C be ring homomorphisms and let h: A ->- B 0A C be defined by hex) = I(x) 0 g(x). Let X, Y, Z, T be the prime spectra of A, B, C, B 0A C respectively. Then h·(T) = I· Y n g·(Z). . [Let.p E X, and let k = k(.p) be the residue field at.p. By Exercise 21, the fiber h·-l(.p) is the spectrum of (B 0A C) 0A k ~ (B 0 Ak) 0k (C 0Ak). Hence .pEh·(T) <=> (B 0Ak) 0k (C 0 Ak) ;j: 0 <::> B 0 Ak i: 0 and C 0Ak i: 0..". l' Ef*(Y) ng*(Z).]

26. Let (Ba, gap) be a direct system of rings and B the direct limit. For each a, let la: A -+ Ba be a ring homomorphism such that gaP ola = Ip whenever a ~ f3 (i.e. the Ba form a direct system of A-algebras). The la induce I: A -+ B. Show that

1·(Spec (B») = n la·(Spec (Ba»). a

[Let l' E Spec (A). Then f* -l(l') is the spectrum of

(since tensor products commute with direct limits: Chapter 2, Exercise 20). By Exercise 21 of Chapter 2 it follows that I·-I(l') = 0 if and only if Ba 0A k(p) = 0 for some IX, i.e., if and only if la· -l(l') = 0.]

27. i) Let la: A ->- Ba be any family of A-algebras and let I: A -+ B be their tensor product over A (Chapter 2, Exercise 23). Then

f*(Spec (B») = n Ia*{Spec (Ba»). a

[Use Examples 25 and 26.] ii) Let fa: A -+ Ba be any finite family of A-algebras and let B = Da Ba.

Define/: A -+ B by I(x) = (faCx»). Thenf*(Spec (B») = Uala·(Spec(Ba»). iii) Hence the subsets of X = Spec (A) of the formf*(Spec (B»), where I: A ->- B

is a ring homomorphism, satisfy the axioms for closed sets in a topological space. The associated topology is the constructible topology on X. It is finer than the Zariski topology (i.e., there are more open sets, or equivalently more closed sets).

iv) Let Xc denote the set X endowed with the constructible topology. Show that Xc is quasi-compact.

28. (Continuation of Exercise 27.) i) For each g E A, the set Xg (Chapter 1, Exercise 17) is both open and closed

in the constructible topology. ii) Let C' denote the smallest topology on X for which the sets Xg are both open

and closed, and let XCI denote the set X endowed with this topology. Show that Xc' is Hausdorff.

iii) Deduce that the identity mapping Xc ->- Xc' is a homeomorphism. Hence a subset E of X is of the formf*(Spec (B») for some/: A -+ B if and only if it is closed in the topology C'.

iv) The topological space Xc is compact, Hausdorff and totally disconnected.

EXERCISES 49

29. Let f: A ->- B be a ring homomorphism. Show that f*: Spec (B) ->- Spec (A) is a continuous closed mapping (i.e., maps closed sets to closed sets) for the con­structible topology.

30. Show that the Zariski topology and the constructible topology on Spec (A) are the same if and only if A/'R is absolutely flat (where 9C is the nilradical of A). [Use Exercise 11.]

4

Primary Decomposition

The decomposition of an ideal into primary ideals is a traditional pillar of ideal theory. It provides the algebraic foundation for decomposing an algebraic variety into its irreducible components-although it is only fair to point out that the algebraic picture is more complicated than naive geometry would suggest. From another point of view primary decomposition provides a gen­eralization of the factorization of an integer as a product of prime-powers. In the modern treatment, with its emphasis on localization, primary decomposition is no longer such a central tool in the theory. It is still, however, of interest in itself and in this chapter we establish the classical uniqueness theorems.

The prototypes of commutative rings are Z and the ring of polynomials k[Xb ... , x,,] where k is a field; both these are unique factorization domains. This is not true of arbitrary commutative rings, even if they are integral domains (the classical example is the ring Z[V - 5], in which the element 6 has two essentially distinct factorizations, 2·3 and (1 + v::5)(l - v=5»). However, there is a generalized form of "unique factorization" of ideals (not of elements) in a wide class of rings (the Noetherian rings).

A prime ideal in a ring A is in some sense a generalization of a prime num­ber. The corresponding generalization of a power of a prime number is a primary ideal. An ideal q in a ring A is primary if q :F A and if

xy E q ~ either x E q or y" E q for some n > O.

In other words,

q is primary <:> A/q :F 0 and every zero-divisor in A/q is nilpotent.

Clearly every prime ideal is primary. Also the contraction of a primary ideal is primary, for iff: A ~ B and if q is a primary ideal in B, then A/qC is isomorphic to a subring of B/q.

Proposition 4.1. Let q be a primary ideal in a ring A. Then r( q) is the smallest prime ideal containing q.

Proof By (1.8) iUs enough to show that.p = r(q) is prime. Let xy E r(q), then (xy)m E q for some m > 0, and therefore either xm E q or ymn E q for some n >~; i.e., either x E r(q) or y E r(q). _

50

PRIMARY DECOMPOSmON 51

If V = r(q), then q is said to be 'p-primary.

Examples. 1) The primary ideals in Z are (0) and (PR), where p is prime. For these are the only ideals in Z with prime radical, and it is immediately checked that they are primary.

2) Let A = k[x, y], q = (x, y2). Then A/q ~ k[y]/(y2), in which the zero­divisors are all the multiples of y, hence are nilpotent. Hence q is primary, and its radical V is (x, y). We have V2 c q C V (strict inclusions), so that a primary ideal is not necessarily a prime-power.

3) Conversely, a prime power VR is not necessarily primary, although its radical is the prime ideal V. For example, let A = k[x, y, z]/(xy - Z2) and let X, y, z denote the images of x, y, z respectively in A. Then V = (x, z) is prime (since A/V ~ k[y], an integral domain); we have xy = Z2 E V2 but x ~ V2 and y ~ r(v2) = V; hence V2 is not primary. However, there is the following result:

Proposition 4.2. If rea) is maximal, then a is primary. In particular, the powers of a maximal ideal mare m-primary.

Proof Let rea) = m. The image ofm in A/a is the nilradical of A/a, hence A/a has only one prime ideal, by (1.8). Hence every element of A/a is either a unit or nilpotent, and so every zero-divisor in A/a is nilpotent. _

We are ioing to study presentations of an ideal as an intersection of primary ideals. First, a couple of lemmas:

Lemma 4.3. If qj (l ~ i ~ n) are 'p-primary, then q = nf-l qj is v-primary. Proof r(q) = r(nf.l qj) = n r(qj) = v. Let xy E q, Y ~ q. Then for some i we have xy E ql and y ~ qb hence x E V, since qj is primary. _

Lemma 4.4. Let q be a v-primary ideal, x an element of A. Then

i) if x E q then (q:x) = (1);

ii) ifx ~ q then (q~x) is v-primary, and therefore r(q:x) = V; iii) if x ~ V then (q:x) = q.

Proof i) and iii) follow immediately from the definitions.

ii): if y E (q:x) then xy E q, hence (as x ~ q) we have y E 'po Hence q s;; (q:x) s; V; taking radicals, we get r(q:x) = V. Let yz E (q:x) with y ~ V; then xyz E q, hence xz E q, hence Z E (q :x). _

A primary decomposition of an ideal a in A is an expression of a as a finite intersection of primary ideals, say

(1)

(In general such a primary decomposition need not exist; in this chapter we shall restrict our attention to ideals which have a primary decomposition.) If more-

52 PRIMARY DECOMPOSITION

over (i) the r(ql) are all distinct, and (ii) we have qj ~ (k .. i qi (1 ~ i ~ n) the primary decomposition (1) is said to be minimal (or irredundant, or reduced, or normal, ... ). By (4.3) we can achieve (i) and then we can omit any superfluous terms to achieve (ii); thus any primary decomposition can be reduced to a minimal one. We shall say that a is decomposable ifit has a primary decomposi­tion.

Theorem 4.5. (1st uniqueness theorem). Let a be a decomposable ideal and let a = (I~-l ql be a minimal primary decomposition of a. Let.pl = r(ql) (1 ~ i ~ n). Then the .pI are precisely the prime ideals which occur in the set of ideals r(a:x) (x E A), and hence are independent of the particular de­composition of a.

Proof. For any x E A we have (a:x) = (n ql:X) = (I (ql:X), hence r(a:x) = (I~-l r(ql:x) = (\.dqf.pi by (4.4). Suppose r(a:x) is prime; then by (1.11) we have r(a:x) = .pi for somej. Hence every prime ideal of the form r(a:x) is one of the .pi. Conversely, for each i there exists Xl ¢= qh XI E (li"l q" since the de­composition is minimal; and we have r(a:xl) = .pl. •

Remarks. 1) The above proof, coupled with the last part of (4.4), shows that for each i there exists Xl in A such that (a: Xl) is .pI-primary.

2) Considering A/a as an A-module, (4.5) is equivalent to saying that the.pl are precisely the prime ideals which occur as radicals of annihilators of elements of A/a.

Example. Let a = (X2, xy) in A = k[x, y]. Then a = .p1 (l.p~ where.p1 = (x), .p2 = (x, y). The ideal .p~ is primary by (4.2). So the prime ideals are .ph .p2. In this example.p1 c .p2; we have rea) = .p1 (l .p2 = .ph but a is not a primary ideal.

The prime ideals.pl in (4.5) are said to belong to a, or to be associated with a. The ideal a is primary if and only if it· has only one associated prime ideal. The minimal elements of the set {.ph ... , .p,,} are called the minimal or isolated prime ideals belonging to a. The others are called embedded prime ideals. In the example above, .p2 = (x, y) is embedded.

Proposition 4.6. Let a be a decomposable ideal. Then any prime ideal .p ;;2 a contains a minimal prime ideal belonging to a, and thus the minimal prime ideals of a are precisely the minimal elements in the set of all prime ideals containing a.

Proof. If .p ;;2 a = (I~-l qh then .p = r(.p) ;;2 n r(ql) = (I .pl. Hence by (1.11) we have.p ;;2 .pI for some i; hence.p contains a minimal prime ideal of a. •

Remarks. 1) The names isolated and embedded come from geometry. Thus if A = k[Xh .. . , x,,] where k is a field, the ideal a gives rise to a variety X s; k" (see Chapter 1, Exercise 25). The minimal primes .pI correspond to the irre­ducible components of X, and the embedded primes correspond to subvarieties

PRIMARY DECOMPOSITION 53

of these, i.e., varieties embedded in the irreducible components. Thus in the example before (4.6) the variety defined by a is the line x = 0, and the embedded ideal -1'2 = (x, y) corresponds to the origin (0, 0).

2) It is not true that all the primary components are independent of the decomposition. For example (x2, xy) = (x) (') (x, y)2 = (x) (') (x2 , y) are two distinct minimal primary decompositions. However, there are some uniqueness properties: see (4.10).

Proposition 4.7. Let a be a decomposable ideal, let a = nf-1 ql be a minimal primary decomposition, and let r( ql) = -1'1' Then

" U -1'1 = {x E A:(a:x) ¥- o}. 1-1

In particular, if the zero ideal is decomposable, the set D of zero-divisors of A is the union of the prime ideals belonging to O.

Proof If a is decomposable, then 0 is decomposable in A/a: namely 0 = n qj where ill is the image of ql in A/a, and is primary. Hence it. is enough to prove the last statement of (4.7). By (1.15) we have D = UHO r(O:x); from the proof of (4.5), we have r(O:x) = nX~qJ -I'j S;; -I'j for some j, hence D s;; Uf-1 -1'1' But also from (4.5) each -1'1 is of the formr(O:x) for some x E A, hence U .);11 S;; D. _

Thus (the zero ideal being decomposable) D = set of zero-divisors

= U of all prime ideals belonging to 0; s:n = set of nilpotent elements

= n of all minimal primes belonging to O.

Next we investigate the behavior of primary ideals under localization.

Proposition 4.8. Let S be a multiplicatively closed subset of A, and let q be a -I'-primary ideal.

i) If S (') -I' ¥- 0, then S-lq = S-lA.

ii) If S (') -I' = 0, then S -lq is S -l-1'-primary and its contraction in A is q. Hence primary ideals correspond to primary ideals in the correspondence

(3.11) between ideals in S-lA and contracted ideals in A.

Proof i) If s E S n -1', then s" E S n q for some n > 0; hence S -lq contains sn/l, which is a unit in S-lA.

ii) If S (') -I' = 0, then s E S and as E q imply a E q, hence q'c = q by (3.11). Also from (3.11) we have r(q·) = r(S-lq) = S-lr(q) = S-l.);1. The verification that S -lq is primary is straightforward. Finally, the contraction of a primary ideal is primary. _

For any ideal 0 and any multiplicatively closed subset S in A, the contraction in A of the ideal S -10 is denoted by S( 0).

54 PRIMARY DECOMPOSITION

Proposition 4.9. Let S be a multiplicatively closed subset of A and let a be a decomposable ideal. Let a = r!r-l ql be a minimal primary decomposition of a. Let ~I = r( ql) and suppose the qj numbered so that S meets ~m + h •.. , ~"

but not .\h, ... , .pm. Then

m S-10 = (') S-1 qh

1=1

and these are minimal primary decompositions.

Proof S-1a = (')?-1 S-1ql by (3.11) = nra 1 S-1ql by (4.8), and S-1ql is S-1.pI-primary for i = 1, ... , m. Since the .pI are distinct, so are the S-1.p1 (1 ~ i ~ m), hence we have a minimal primary decomposition. Contracting both sides, we get

by (4.8) again. _

A set ~ of prime ideals belonging to a is said to be isolated if it satisfies the following condition: if .p' is a prime ideal belonging to a and .p' s;; .p for some ~ E ~, then .p' E ~.

Let ~ be an isolated set of prime ideals belonging to a, and let S = A -U~Ell %'. Then S is multiplicatively closed and, for any prime ideal .p' belonging to a, we have

.p' E ~ => .p' (\ S = 125;

.p' ¢ ~ => ~' $ U .p (by (1.11» => .p' n S;6 125. pel:

Hence, from (4.9), we deduce

Theorem 4.10. (2nd uniqueness theorem). Let a be a decomposable ideal, let a = (')?-1 ql be a minimal primary decomposition of a, and let {.ph' ... , .pI .. } be an isolated set of prime ideals of a. Then ql1 n· .. n ql .. is independent of the decomposition.

In particular:

Corollary 4.11. The isolated primary components (i.e., the primary com­ponents ql corresponding to minimal prime ideals .pI) are uniquely determined by a. •

Proofof(4.10). We have qj1 (\···n ql .. = Sea) where S = A - .p11 u···u .pI .. , hence depends only on a (since the ~j depend only on al... _

Remark. On the other hand, the embedded primary components are in general not uniquely determined by a. If A is a Noetherian ring, there are in fact infinitely many choices for each embedded component (see Chapter 8, Exercise 1).

EXERCISES 55

EXEROSES

1. If an ideal a has a primary decomposition, then Spec (A/a) has only finitely many irreducible components.

2. If a = r(a), then a has no embedded prime ideals.

3. If A is absolutely flat, every primary ideal is maximal.

4. In the polynomial ring Z[t], the ideal m = (2. t) is maximal and the ideal q = (4,1) is m-primary, but is not a power of m.

5. In the polynomial ring K[x, y, z] where K is a field and x, y, z are independent indeterminates, let.pl = (x, y),.p2 = (x, z), m = (x, y, z); .pI and .p2 are prime, and m is maximal. Let a = .pl.p2. Show that a = .pI n .p2 n m 2 is a reduced primary decomposition of a. Which components are isolated and which are embedded?

6. Let X be an infinite compact Hausdorff space, C(X) the ring of real-valued continuous fUnctions on X (Chapter 1, Exercise 26). Is the zero ideal de­composable in this ring?

7. Let A be a ring and let A [x] denote the ring of polynomials in one indeterminate over A. For each ideal a of A, let a[x] denote the set of all polynomials in A[x] with coefficients in a.

i) a[x] is the extension of a to A[x]. ii) If.p is a prime ideal in A, then .p[x] is a prime ideal in A[x]. iii) If q is a .p-primary ideal in A, then q[x] is a .p[x]-primary ideal in A[x].

[Use Chapter I, Exercise 2.] iv) If a = (lfsl ql is a minimal primary decomposition in A, then a[x] =

nr.l ql[x] is a minimal primary decompOSition in A[x]. v) If.p is a minimal prime ideal of a, then .p[x] is a minimal prime ideal of a[x].

8. Let k be a field. Show that in the polynomial ring k[Xl, ... , x,,] the ideals .pI = (Xl, ... , XI) (1 .;;; i .;;; n) are prime and all their powers are primary. [Use Exercise 7.]

9. In a ring A, let D(A) denote the set of prime ideals .p which satisfy the following condition: there exists a E A such that .p is minimal in the set of prime ideals containing (0: a). Show that X E A is a zero divisor ¢> x E .p for some .p E D(A).

Let S be a multiplicatively closed subset of A, and identify Spec (S-IA) with its image in Spec (A) (Chapter 3, Exercise 21). Show that

D(S-IA) = D(A) n Spec (S-IA).

If the zero ideal has a primary decomposition, show that D(A) is the set of associated prime ideals of O.

10. For any prime ideal.p in a ring A, let S,,(O) denote the kernel of the homo-morphism A -+ A". Prove that

i) S,,(O) S; .p. ii) r(S,,(O») = .p ¢> .p is a minimal prime ideal of A.

iii) If.p ;2 .p', then S,,(O) S; s", (0). iv) n"&D(A) S,,(O) = 0, where D(A) is defined in Exercise 9.

3 +r.C.A.

56 PRIMARY DECOMPOSITION

11. If.p is a minimal prime ideal of a ring A, show that S~(O) (Exercise 10) is the smallest .p-primary ideal.

Let a be the intersection of the ideals Sll(O) as .p runs through the minimal prime ideals of A. Show that a is contained in the nilradical of A.

Suppose that the zerO ideal is decomposable. Prove that a = 0 if and only if every prime ideal of 0 is iSOlated.

12. Let A be a ring, S a multiplicatively closed subset of A. For any ideal a, let Sea) denote the contraction of S -la in A. The ideal Sea) is called the saturation of a with respect to S. Prove that i) Sea) (') S(b) = sea (') b)

ii) S(r(a» = r(S(a» iii) Sea) = (1) ¢> a meets S iv) Sl(S2(a» = (SlS2)(a). If a has a primary decomposition, prove that the set of ideals Sea) (where S runs through all multiplicatively closed subsets of A) is finite.

13. Let A be a ring and .p a prime ideal of A. The nth symbolic power oJ.p is defined to be the ideal (in the notation of Exercise 12)

where Sp = A -.p. Show that i) .p(n) is a .p-primary ideal;

.p(n) = Sp(.pn)

ii) if.pn has a primary decomposition, then .p(n) is its .p-primary component; iii) if .p(Ift).p(n) has a primary decomposition, then .p(m + n) is its .p-primary compo­

nent; iv) .p(n) = .pn ¢> .p(n) is .p-primary.

14. Let a be a decomposable ideal in a ring A and let .p be a maximal element of the set of ideals (a :x), where x E A and x¢: a. Show that.p is a prime ideal belonging to a.

15. Let a be a decomposable id~al in a ring A, let 1: be an isolated set of prime ideals belonging to a, and let ql: be the intersection of the corresponding primary components. LetJbe an element of A such that, for each prime ideal.p belonging to a, we have IE.p ¢>.p ¢: 1:, and let Sf be the set of all powers of f Show that ql: = Sf(a) = (a:Jn) for all large n.

16. If A is a ring in which every ideal has a primary decomposition, show that every ring of fractions S -1 A has the same property.

17. Let A be a ring with the following property. (LI) For every ideal a i:- (1) in A and every prime ideal.p, there exists x¢: .p such that Sp(a) = (a:x), where S~ = A - .p.

Then every ideal in A is an intersection of (possibly infinitely many) primary ideals. [Let a be an ideal i:- (1) in A, and let.pl be a minimal element of the set of prime ideals containing a. Then ql = Sll,(a) is .pl-primary (by Exercise 11), and ql = (a:x) for some x ¢: .pl. Show that a = ql (') (a + (x».

Now let al be a maximal element of the set of ideals b ;2 a such that ql (') b = a, and choose al so that x E a17 and therefore al $ .p" Repeat the

EXERCISES 57

construction starting with at. and so on. At the nth stage we have a = ql /l ... /l qn /l an where the qj are primary ideals, an is maximal among the ideals b con­taining an -1 = an /l qn such that a = ql /l ... /l qn /l b, and an $ Pn. If at any stage we have an = (1), the process stops, and a is a finite intersection of primary ideals. If not, continue by transfinite induction, observing that each an strictly contains an -1.]

18. Consider the following condition on a ring A: (L2) Given an ideal a and a descending chain Sl ;2 S2 ;2 ... ;2 Sn;2··· of multiplicatively closed subsets of A, there exists an integer n such that Sn(a) Sn+l(a) = .... Prove that the following are equivalent: i) Every ideal in A has a primary decomposition;

ii) A satisfies (Ll) and (L2). [For i) =!> ii), use Exercises 12 and 15. For ii) =;- i) show, with the notation of the proof of Exercise 17, that if Sn = SPl /l ... /l SP. then Sn meets an, hence Sn(an) = (1), and therefore Sn(a) = ql /l ... /l qn. Now use (L2) to show that the construction must terminate after a finite number of steps.]

19. Let A be a ring and P a prime ideal of A. Show that every p-primary ideal contains Sp(O), the kernel of the canonical homomorphism A -+ Ap.

Suppose that A satisfies the following condition: for every prime ideal p, the intersection of all p-primary ideals of A is equal to Sp(O). (Noetherian rings satisfy this condition: see Chapter 10.) Let PI. ... , Pn be distinct prime ideals, none of which is a minimal prime ideal of A. Then there exists an ideal a in A whose associated prime ideals are Pl> ... , Pn. [Proof by induction on n. The case n = 1 is trivial (take a = lJl). Suppose n > 1 and let Pn be maximal in the set {Pl, ... , Pn}. By the inductive hypothesis there exists an ideal b and a minimal primary decomposition lJ = ql /l ... /l qn-l, where each qj is pj-primary. If lJ £; SP.(O), let P be a minimal prime ideal of A contained in Pn. Then SP.(O) £; Sp(O), hence lJ £; Sp(O). Taking radicals and using Exercise 10, we have Pl /l···/lPn_l £; p, hence some pj £; p, hence pj = P since P is minimal. This is a contradiction since no pj is minimal. Hence b $ SP.(O) and therefore there exists a Pn-primary ideal qn such that b $ qn. Show that a = ql /l ... /l qn has the required properties.]

Primary decomposition of modules Practically the whole of this chapter can be transposed to the context of

modules over a ring A. The following exercises indicate how this is done.

20. Let M be a fixed A-module, N a submodule of M. The radical of N in Mis defined to be

rM(N) = {x E A:xqM £; N for some q > O}.

Show that rM(N) = r(N:M) = r(Ann (M/N». In particular, 'M(N) is an ideal.

State and prove the formulas for rM analogous to (1.13).

21. An element x E A defines an endomorphism ~x of M, namely m f-+ xm. The element x is said to be a zero-divisor (resp. nilpotent) in M if ~x is not injective

58 PRIMARY DECOMPOSITION

(resp. is nilpotent). A submodule Q of M is primary in M if Q i:- M and every zero-divisor in MI Q is nilpotent.

Show that if Q is primary in M, then (Q: M) is a primary ideal and hence rM(Q) is a prime ideal~. We say that Q is ~-primary (in M).

Prove the analogues of (4.3) and (4.4).

22. A primary decomposition of N in M is a representation of N as an intersection

N = Ql n···n Qn

of primary submodules of M; it is a minimal primary decomposition if the ideals ~I = rM(QI) are all distinct and if none of the components QI can be omitted from the intersection, that is if QI ~ (li"l QJ (1 .:;; i .:;; n).

Prove the analogue of (4.5), that the prime ideals ~I depend only on N (and M). They are called the prime ideals belonging to N in M. Show that they are also the prime ideals belonging to 0 in MIN.

23. State and prove the analogues of (4.6)-{4.11) inclusive. (There is no loss of generality in taking N = 0.)

5

Integral Dependence and

Valuations

In classical algebraic geometry curves were frequently studied by projecting them onto a line and regarding the curve as a (ramified) covering of the line. This is quite analogous to the relationship between a number field and the rational field-or rather between their rings of integers-and the common algebraic feature is the notion of integral dependence. In this chapter we prove a number of results about integral dependence. In particular we prove the theo­rems of Cohen-Seidenberg (the "going-up" and "going-down" theorems) concerning prime ideals in an integral extension. In the exercises at the end we discuss the algebro-geometric situation and in particular the Normalization Lemma.

We also give a brief treatment of valuations.

INTEGRAL DEPENDENCE

Let B be a ring, A a sub ring of B (so that 1 E A). An element x of B is said to be integral over A if x is a root of a monic polynomial with coefficients in A, that is if x satisfies an equation of the form

(1)

where the at are elements of A. Clearly every element of A is integral over A.

Example 5.0. A = Z, B = Q. If a rational number x = rls is integral over Z, where r, s have no common factor, we have from (1)

r" + aIr"-ls + ... + a"s" = 0

the at being rational integers. Hence s divides r", hence s = ± 1, hence x E Z.

Proposition 5.1. The following are equivalent:

i) x E B is integral over A;

ii) A[x] is afinitely generated A-module;

iii) A [x] is contained in a subring C of B such that C is a finitely generated A-module; .

59

60 INTEGRAL DEPENDENCE AND VALUATIONS

iv) There exists a faithful A [x]-module M which is finitely generated as an A-module.

Proof. i) => ii). From (1) we have

for all r ;?: 0; hence, by induction, all positive powers of x lie in the A-module generated by l,x, ... ,x"-1. Hence A[x] is generated (as an A-module) by 1, x, . . .• x" - 1.

ii) => iii). Take C = A[x].

iii) => iv). Take M = C, which is a faithful A[x]-module (since yC = 0 =>

y·l = 0).

iv) => i). This follows from (2.4): take'" to be multiplication by x, and a = A (we have xM S;; M since M is an A[x]-module); since M is faithful, we have x" + a1Xn-1 + ... + a" = 0 for suitable a l EA. _

Corollary 5.2. Let XI (1 ~ i ~ n) be elements of B, each integral over A. Then the ring A[X1, ... , x,,] is afinitely-generated A-module.

Proof By induction on n. The case n = 1 is part of (5.1). Assume n > 1, let Ar = A[Xb ... , xr]; then by the inductive hypothesis A n- 1 is a finitely generated A-module. An = A"-1[X,,] is a finitely generated An _ 1-modu1e (by the case n = 1, since x" is integral over An -1)' Hence by (2.16) An is finitely generated as an A-module. _

Corollary 5.3. The set C of elements of B which are integral over A is a subring of B containing A.

Proof. If x, y E C then A [x, y] is a finitely generated A-module by (5.2). Hence x ± y and xy are integral over A, by iii) of (5.1). _

The ring C in (5.3) is called the integral closure of A in B. If C = A, then A is said to be integrally closed in B. If C = B, the ring B is said to be integral over A.

Remark. Let f: A -+ B be a ring homomorphism, so that B is an A-algebra. Then f is said to be integral, and B is said to be an integral A-algebra, if B is integral over its subringf(A). In this terminology, the above results show that

finite type + integral = finite.

Corollary 5.4. If A S;; B S;; C are rings and if B is integral over A, and C is integral over B, then C is integral over A (transitivity of integral dependence).

Proof Let x E C, then we have an equation

(bl E B).

The ring B' = A[b1 • ... , b,,] is a finitely generated A-module by (5.2), and B'[x] is a finitely generated B'-module (since x is integral over B'). Hence B'[x] is a

THE GOING-UP THEOREM 61

finitely generated A-module by (2.16) and therefore x is integral over A by iii) of (5.1). •

Corollary 5.5. Let A £; B be rings and let C be the integral closure of A in B. Then C is integrally closed in B.

Proof Let x E B be integral over C. By (5.4) x is integral over A, hence XEC .•

The next proposition shows that integral dependence is preserved on passing to quotients and to rings of fractions:

Proposition 5.6. Let A £; B be rings, B integral over A.

i) Ifb is an ideal of B and a = be = An b, then Bib is integral over Ala.

ii) If S is a multiplicatively closed subset of A, then S - 1 B is integral over S-lA.

Proof i) If x E B we have, say, xn + a1xn -1 + ... + an = 0, with al E A. Reduce this equation mod. b.

ii) Let xis E S-lB(x E B, s E S). Then the equation above gives

(xls)n + (a1Is)(xls)n-1 + ... + anlsn = 0

which shows that xis is integral over S-l A. •

THE GOING-UP THEOREM

Proposition 5.7. Let A £; B be integral domains, B integral over A. Then B is a field if and only if A is a field.

Proof Suppose A is a field; let y E B, y # O. Let

be an equation of integral dependence for y of smallest possible degree. Since B is an integral domain we have a" # 0,hencey-1 = _a;1(yn-1 + a1yn-2 + ... + a" -1) E B. Hence B is a field.

Conversely, suppose B is a field; let x E A, x # O. Then x- 1 E B, hence is integral over A, so that we have an equation

x - m + a~ x - m + 1 + ... + a~ = 0 (a; E A).

It follows that X-I = -(a~ + a;x + ... + a~xm-1) E A, hence A is a field. •

Corollary 5.g. Let A £; B be rings, B integral over A; let q be a prime ideal of B and let ~ = qe = q n A. Then q is maximal if and only if ~ is maximal.

Proof By (5.6), Blq is integral over A/~, and both these rings are integral domains. Now use (5.7). •

Corollary 5.9. Let A £; B be rings, B integral over A; let q, q' be prime ideals of B such that q £; q' and qe = q'e = ~ say. Then q = q'.

62 INTEGRAL DEPENDENCE AND VALUATIONS

Proof By (5.6), Bll is integral over All' Let m be the extension of.p in All and let n, n' be the extensions of q, q' respectively in Bll . Then m is the maximal ideal of All; n s;; n', and nC = n'C = m. By (5.8) it follows that n, n' are maximal, hence n = n', hence by (3.l1 )(iv) q = q'. •

Theorem 5.10. Let A S;; B be rings, B integral over A, and let .p be a prime ideal of A. Then there exists a prime ideal q of B such that q n A = .p.

Proof By (5.6), Bll is integral over All' and the diagram

A-+B at t,B All-+Bll

(in which the horizontal arrows are injections) is commutative. Let n be a maxi­mal ideal of Bll ; then m =. n n All is maximal by (5.8), hence is the unique maximal ideal of the local ring All' If q = f3 -1( n), then q is prime and we have q n A = a:- 1(m) =.p .•

Theorem 5.11. ("Going-up theorem"). Let A S;; B be rings, B integral over A; let .pl S;; ••• S;; .p" be a chain of prime ideals of A and q s;; ... S;; qm (m < n) a chain of prime ideals of B such that ql n A = .pI (1 ~ i ~ m). Then the chain q1 s;; ... S;; qm can be extended to a chain ql s;; •.• s;; q" such that qj n A = .pdor 1 ~ i ~ n.

Proof By induction we reduce immediately to the case m = 1, n = 2. Let A = A/.ph E = B/ql; then A S;; E, and E is integral over A by (5.6). Hence, by (5.10), there exists a prime ideal 112 of E such that 112 n A = i32' the image of.p2 in A. Lift back 112 to B and we have a prime ideal q2 with the required pro­perties. •

INTEGRALLY CLOSED INTEGRAL DOMAINS. THE GOING-DOWN THEOREM

Proposition (5.6)(ii) can be sharpened:

Proposition 5.12. Let A S;; B be rings, C the integral closure of A in B. Let S be a multiplicatively closed subset of A. Then S -1e is the integral closure of S - 1 A in S - lB.

Proof By (5.6), s-le is integral over S-lA. Conversely, if blsES- 1B is integral over S-lA, then we have an equation of the form

(bls)" + (al/s1)(bls)n-l + ... + anlsn = 0

where al E A, Sl E S (1 ~ i ~ n). Let t = S1' .• s" and multiply this equation by (st? throughout. Then it becomes an equation of integral dependence for bt over A. Hence bt E e and therefore bls = bllst E S -le. •

An integral domain is said to be integrally closed (without qualification) if it is integrally closed in its field of fractions. For example, Z is integrally

INTEGRALLY CLOSED INTEGRAL DOMAINS. GOING-DOWN THEOREM 63

closed (see (S.O»). The same argument shows that any unique factorization domain is integrally closed. In particular, a polynomial ring k[Xlo ... , x,,] over a field is integrally closed.

Integral closure is a local property:

Proposition 5.13. Let A be an integral domain. Then the following are equivalent:

i) A is integrally closed;

ii) AlJ is integrally closed, for each prime ideal .p;

iii) Am is integrally closed, for each maximal ideal m.

Proof Let Kbe the field of fractions of A, let C be the integral closure of A in K, and let f: A -+ C be the identity mapping of A into C. Then A is integrally closed <:> f is surjective, and by (S.12) AlJ (resp. Am) is integrally closed <:> fll (resp. fm) is surjective. Now use (3.9). _

Let A S;; B be rings and let a be an ideal of A. An element of B is said to be integral over a if it satisfies an equation of integral dependence over A in which all the coefficients lie in a. The integral closure of a in B is the set of all elements of B which are integral over a.

Lemma 5.14. Let C be the integral closure of A in B and let as denote the extension of a in C. Then the integral closure of a in B is the radical of a' (and is therefore closed under addition and multiplication).

Proof If x E B is integral over a, we have an equation of the form

with al> ••. , a" in a. Hence x E C and x" E as, that is x E rea'). Conversely, if x E r(a') then x" = L: ajXj for some n > 0, where the a l are elements of a and the Xl are elements of C. Since each Xl is integral over A it follows from (S.2) that M = A[xl>"" x,,] is a finitely generated A-module, and we have x"M S;; aM. Hence by (2.4) (taking '" there to be multiplication by x") we see that x" is integral over a, hence x is integral over a. _

Proposition 5.15. Let A S;; B be integral domains, A integrally closed, and let x E B be integral over an ideal a of A. Then x is algebraiC over the field of fractions K of A, and if its minimal polynomial over K is t" + a1t,,-1 + ... + a", then al> ... , a" lie in rea).

Proof Clearly x is algebraic over K. Let L be an extension field of K which contains all the conjugates Xl> .•• , x" of x. Each Xl satisfies the same equation of integral dependence as x does, hence each Xl is integral over a. The co­efficients of the minimal polynomial of x over K are polynomials in the Xl> hence by (S.14) are integral over a. Since A is integrally closed, they must lie in rea), by (S.14) again. _

3*

64 INTEGRAL DEPENDENCE AND VALUATIONS

Theorem 5.16. ("Going-down theorem"). Let A S;; B be integral domains, A integrally closed, B integral over A. Let P1 ;;2 .•. ;;2 Pn be a chain of prime ideals of A, and let q1 ;;2 ••• 2 qm (m < n) be a chain of prime ideals of B such that ql n A = PI (1 ~ i ~ m). Then the chain q1 ;;2 •.. 2 qm can be extended to a chain q1 ;;2 ••. ;;2 qn such that ql n A = PI (1 ~ i ~ n).

Proof As in (5.11) we reduce immediately to the case m = 1, n = 2. Then we have to show that P2 is the contraction of a prime ideal in the ring Bq1 , or equivalently (3.16) that Bq1 P2 n A = P2'

Every x E Bq1P2 is of the form yls, where y E Bp2 and s E B - q1' By (5.14), y is integral over P2, and hence by (5.15) its minimal equation over K, the field of fractions of A, is of the form

(1)

with U1, ... , UT in P2' Now suppose that x E Bq1 P2 n A. Then s = yx- 1 with X-I E K, so that the

minimal equation for s over K is obtained by dividing (1) by x r, and is therefore,

say, sr + V1S T

-1 + ... + VT = 0

where VI = ullxl. Consequently

(2)

XlVI = Ul E P2 (1 ~ i ~ r). (3)

But s is integral over A, hence by (5.15) (with a = (1») each VI is in A. Suppose x ¢= P2' Then (3) shows that each VI E P2, hence (2) shows that ST E Bp2 S;; Bp1 S;; q1, and therefore s E q1' which is a contradiction. Hence x E P2 and therefore Bq1P2 n A = P2 as required. -

The proof of the next proposition assumes some standard facts from field theory.

Proposition 5.17. Let A be an integrally closed domain, K itsfield of fractions, L afinite separable algebraiC extension of K, B the integral closure of A in L. Then there exists a basis V1> ... , Vn of Lover K such that B S;; 2)-1 AVj.

Proof If V is any element of L, then V is algebraic over K and therefore satisfies an equation of the form

CloVT + a1vr-1 + ... + an = 0 (al E A).

Multiplying this equation by ao-1, we see that aov = u is integral over A, and hence is in B. Thus, given any basis of Lover K we may multiply the basis ele­ments by suitable elements of A to get a basis U1> ... , Un such that each Ul E B.

Let T denote trace (from L to K). Since LIK is separable, the bilinear form (x, y) 1-+ T(xy) on L (considered as a vector space over K) is non-degenerate, and hence we have a dual basis V1, ... , vn. of Lover K, defined by T(ulv,) = olj. Let x E B, say x = L, XiV, (x, E K). We have XUI E B (since Ul E B) and therefore T(XUI) E A by (5.15) (for the trace of an element is a multiple of one of the co­efficients in the minimal polynomial). But T(xul) = L' T(xjulv,) = L, X,T(UIV,) = Lj Xj 0li = Xi' hence Xl E A. Consequently B S;; Lj AVj. -

VALUATION RINGS 65

VALUATION RINGS

Let B be an integral domain, K its field of fractions. B is a valuation ring of K if, for each x #: 0, either x E B or x- 1 E B (or both).

Proposition 5.18. i) B is a local ring.

ii) If B' is a ring such that B s; B' s; K, then B' is a valuation ring of K.

iii) B is integrally closed (in K).

Proof i) Let m be the set of non-units of B, so that x Em¢> either x = 0 or x -1 ¢ B. If a E B and x E m we have ax Em, for otherwise (ax) - 1 E B and therefore x -1 = a· (ax) -1 E B. Next let x, y be nOn-zero elements of m. Then either xy -1 E B or x- 1y E B. If xy-1 E B then x + y = (1 + xy-1)y E Bm s; m, and similarly if x- 1y E B. Hence m is an ideal and therefore B is a local ring by (1.6).

ii) Clear from the definitions. iii) Let x E K be integral Over B. Then we have, say,

xn + b1xn -1 + ... + bn = 0

with the bl E B. If x E B there is nothing to prove. If not, then x -1 E B, hence x = -(b1 + b2x- 1 + ... + bnx 1- n)EB. 1'1

Let K be a field, n an algebraically closed field. Let ~ be the set of all pairs (A, I), where A is a subring of K andfis a homomorphism of A into n. We partially order the set ~ as follows:

(A,J) ~ (A',J') ¢> A S A' andf'IA = f The conditions of Zorn's lemma are clearly satisfied and therefore the set ~ has at least one maximal element.

Let (B, g) be a maximal element of~. We want to prove that B is a valuation ring of K. The first step in the proof is

Lemma 5.19. B is a local ring andm = Ker (g) is its maximal ideal.

Proof Since g(B) is a subring of a field and therefore an integral domain, the ideal m = Ker (g) is prime. We can extend g to a homomorphism if: Bm -+ n by putting if(b/s) = g(b)/g(s) for all b E B and all s E B - m, since g(s) will not be zero. Since the pair (B, g) is maximal it follows that B = Bm, hence B is a local ring and m is its maximal ideal. _

Lemma 5.20. Let x be a non-zero element of K. Let B[x] be the subring of K generated by x over B, and let m[x] be the extension olm in B[x]. Then either m[x] #- B[x] or m[x- 1] #: B[x- 1].

Proof Suppose that m[x] = B[x)" and m[x- 1] = B[x- 1]. Then we shall have equations

Uo + U1X + ... + umxm = 1 (UI Em)

Vo + V1X-1 + ... + VnX- n = 1 (VjEm)

(1)

(2)

in which we may assume that the degrees m, n are as small as possible. Suppose that m ~ n, and mUltiply (2) through by xn:

(3)

66 INTEGRAL DEPENDENCE AND VALUATIONS

Since Vo Em, it follows from (5.19) that 1 - Vo is a unit in B, and (3) may therefore be written in the form

(Wj Em).

Hence we can replace xm in (1) by W1X .. -1 + ... + w"xm- n, and this contradicts the minimality of the exponent m. •

Theorem 5.21. Let (B, g) be a maximal element of~. Then B is a valuation ring of the field K.

Proof. We have to show that if x i:- 0 is an element of K, then either x E B or X-l E B. By (5.20) we may as well assume that m[x] is not the unit ideal of the ring B' = B[x]. Then m[x] is contained in a maximal ideal m' of B', and we have m' n B = m (because m' n B is a proper ideal of B and contains m). Hence the embedding of B in B' induces an embedding of the field k = Blm in the field k' =

B'/m'; also k' = k[x] where x is the image of x in k', hence x is algebraic over k, and therefore k' is a finite algebraic extension of k.

Now the homomorphism g induces an embedding g of kin n, since by (5.19) m is the kernel of g. Since n is algebraically closed, g can be extended to an em­bedding g' of k' into n. Composing g' with the natural homomorphism B' -- k', we have, say, g': B' ->- n which extends g. Since the pair (B, g) is maximal, it follows that B' = B and therefore x E B. •

Corollary 5.22. Let A be a subring of a field K. Then the integral closure A of A in K is the intersection of all the valuation rings of K which contain A.

Proof. Let B be a valuation ring of K such that A S; B. Since B is integrally closed, by (5.18) iii), it follows that A S; B.

Conversely, let x¢ A. Then x is not in the ring A' = A[x- l ]. Hence X-l is a non-unit in A' and is therefore contained in a maximal ideal m' of A'. Let n be an algebraic closure of the field k' = A'/m'. Then the restriction to A of the natural homomorphism A' ->- k' defines a homomorphism of A into n. By (5.21) this can be extended to some valuation ring B ;2 A. Since x -1 maps to zero, it follows that x¢B .•

Proposition 5.23. Let A s; B be integral domains, B finitely generated over A. Let v be a non-zero element of B. Then there exists u i:- 0 in A with the fol/owing property,' any homomorphism f of A into an algebraically closed field n such that f(u) i:- 0 can be exte"ded to a homomorphism g of B into n such that g(v) i:- O.

Proof. By induction on the number of generators of B over A we reduce immediately to the case where B is generated over A by a single element x.

i) Suppose x is transcendental over A, i.e., that no non-zero polynomial with coefficients in A has x as a root. Let v = aox" + alx"-l + ... + an, and take u = ao. Then iff: A ->- n is such thatf(u) i:- 0, there exists, En such thatf(ao)~ + f(al)~-l + ... + !(an) i:- 0, because n is infinite. Define g: B -- n extendingf by putting g(x) = ,. Then g(v) i:- 0, as required.

ii) Now suppose x is algebraic over A (i.e. over the field of fractions of A). Then so is V-I, because v is a polynomial in x. Hence we have equations of the form

aoX"' + a1xm- 1 + ... + am = 0 (a, E A) (1)

aov- n + a~v1-n + .. , + a;. = 0 (a; E A). (2)

EXERCISES 67

Let u = aoao, and let/: A ->- n be such that/(u) "# o. Then/can be extended, first to a homomorphismh: A[u-1] ->- n (withh(u- 1) = /(U)-I), and then by (S.21) to a homomorphism h: C ->- n, where C is a valuation ring containing A [u -1]. From (1), x is integral over A[u- 1 ], hence by (S.22) x E C, so that C contains B, and in particular v E C. On the other hand, from (2), V-I is integral Over A[u- 1 ], and there­fore by (S.22) again is in C. Therefore v is a unit in C, and hence h(v) "# O. Now take g to be the restriction of h to B. •

Corollary 5.24. Let k be a field and B a finitely generated k-algebra. 1/ B is a field then it is a finite algebraic extension 0/ k.

Proof. Take A = k, v = 1 and n = algebraic closure of k. •

(S.24) is one form of Hilbert's Nullstellensatz. For another proof, see (7.9).

EXERCISES

1. Let/: A -+ B be an integral homomorphism of rings. Show that/·: Spec (B)->­Spec.(A) is a closed mapping, i.e. that it maps closed sets to closed sets. (This is a geometrical equivalent of (S.10).)

2. Let A be a subring of a ring B such that B is integral over A, and let /: A ->- n be a homomorphism of A into an algebraically closed field n. Show that/can be extended to a homomorphism of B into n. [Use (S.10).]

3. Let /: B ->- B' be a homomorphism of A-algebras, and let C be an A-algebra. If f is integral, prove that / 0 1: B 0 A C ->- B' 0 A C is integral. (This includes (S.6) ii) as a special case.)

4. Let A be a subring of a ring B such that B is integral over A. Let n be a maximal ideal of B and let m = n n A be the corresponding maximal ideal of A. Is Bn necessarily integral over Am? [Consider the subring k[x2 - 1] of k[x], where k is a field, and let n = (x - 1). Can the element l/(x + 1) be integral?]

5. Let A S B be rings, B integral over A. i) If x E A is a unit in B then it is a unit in A.

ii) The Jacobson radical of A is the contraction of the Jacobson radical of B.

6. Let Blo ... , Bn be integral A-algebras. Show that n~= 1 Bf is an integral A­algebra.

7. Let A be a subring of a ring B, such that the set B - A is closed under multi­plication. Show that A is integrally closed in B.

8. i) Let A be a subring of an integral domain B, and let C be the integral closure of A in B. Let J, g be monic polynomials in B[x] such that /g E C [x]. Then J, g are in C[x]. [Take a field containing B in which the polynomials J, g

split into linear factors: say / = II (x - 'I), g = II (x - TIJ). Each 'I and each 7)j is a root of /g, hence is integral over C. Hence the coefficients of / and g are integral over C.]

ii) Prove the same result without assuming that B (or A) is an integral domain.

68 INTEGRAL DEPENDENCE AND VALUATIONS

9. Let A be a subring of a ring B and let C be the integral closure of A in B. PrOve that C[x] is the integral closure of A[x] in B[x]. [If fE B[x] is integral over A[x], then

f'" + gJm-1 + ... + gm = 0 (gj E A[xD.

Let r be an integer larger than m and the degrees of g1, ... , gm, and let f1 = f - x T

, so that

or say fr + hdr -1 + ... + hm = 0,

where hm = (xT)m + gl(XT)m-1 + ... + gm E A[x]. Now apply Exercise 8 to the polynomials -fi. andft- 1 + hdr- 2 + ... + hm - 1 .]

10. A ring homomorphismf: A -)0 B is said to have the going-up property (resp. the going-down property) if the conclusion of the going-up theorem (5.11) (resp. the gOing-down theorem (5.16)) holds for B and its sUbringf(A).

Let f*: Spec (B) ~ Spec (A) be the mapping associated with f i) Consider the following three statements:

(a) f* is a closed mapping. (b) fhas the going-up property. (c) Let q be any prime ideal of B and let ~ = qC. Then f*: Spec (Blq) ->­

Spec (AM is surjective. Prove that (a) => (b) ¢> (c). (See also Chapter 6, Exercise 11.)

ii) Consider the following three statements: (a') f* is an open mapping. (b') fhas the going-down property. (c') For any prime ideal q of B, if ~ = qc, thenf*: Spec (Bq) -+ Spec (All) is

surjective. Prove that (a') => (b') ¢> (c'). (See also Chapter 7, Exercise 23.)

[To prove that (a') =!> (c'), observe that Bq is the direct limit of the rings Bt

where t E B - q; hence, by Chapter 3, Exercise 26, we have f*(Spec (Bq») =

(Jd*(Spec (Bt» = (Jd*( Yt ). Since Yt is an open neighborhood of q in Y, and since f* is open, it follows that f*( Yt ) is an open neighborhood of ~ in X and therefore contains Spec (All)']

11. Let f: A -+ B be a flat homomorphism of rings. Then f has the going-down property. [Chapter 3, Exercise 18.]

12. Let G be a finite group of automorphisms of a ring A, and let AG denote the subring of G-invariants, that is of all x E A such that o'(X) = x for all a E G. Prove that A is integral over AG. [If x E A, observe that x is a root of the poly­nomial IIqeG (t - a(x»).]

Let S be a multiplicatively closed subset of A such that O'(S) S S for all a E G, and let sa = SnAG. Show that the action of G on A extends to an action on S-1A, and that (SG)-1AG ~ (S-1A)G.

13. In the situation of Exercise 12, let ~ be a prime ideal of AG, and let P be the set of prime ideals of A whose contraction is~. Show that G acts transitively on P. In particular, P is finite.

EXERCISES 69

[Let .\hP2 E P and let x E Pl. Then II" a(x) E PI nAG = P s; P2, hence a(x) EP2 for some a E G. Deduce that PI is contained in U"eG a(P2), and then apply (1.11) and (5.9).]

14. Let A be an integrally closed domain, K its field of fractions and L a finite normal separable extension of K. Let G be the Galois group of Lover K and let B be the integral closure of A in L. Show that a(B) = B for all a E G, and that A = BG.

15. Let A, K be as in Exercise 14, let L be any finite extension field of K, and let B be the integral closure of A in L. Show that, if P is any prime ideal of A, then the set of prime ideals q of B which contract to P is finite (in other words, that Spec (B) -+ Spec (A) has finite fibers). [Reduce to the two cases (a) L separable over K and (b) L purely inseparable over K. In case (a), embed L in a finite normal separable extension of K, and use Exercises 13 and 14. In case (b), if q is a prime ideal of B such that q n A = p, show that q is the set of all x E B such that x pln

E P for some m ~ 0, where p is the characteristic of K, and hence that Spec (B) -+ Spec (A) is bijective in this case.]

Noether's normalization lemma

16. Let k be a field and let A i:- 0 be a finitely generated k-algebra. Then there exist elements Y10 ••• , Yr E A which are algebraically independent over k and such that A is integral over k[Yl, ... , Yr].

We shall assume that k is infinite. (The result is still true if k is finite, but a different proof is needed.) Let Xl, ... , Xn generate A as a k-algebra. We can renumber the Xj so that Xl> ... , Xr are algebraically independent over k and each of xr+ 1, .•• , Xn is algebraic over k[Xl, ... , x r]. Now proceed by induction on n. If n = r there is nothing to do, so suppose n > r and the result true for n - 1 generators. The generator Xn is algebraic over k[Xl,"" xn-tl, hence there exists a pOlynomial f i:- 0 in n variables such that f(xl> ... , Xn -10 xn) = O. Let F be the homogeneous part of highest degree in f. Since k is infinite, there exist ,110 .•• , An -1 E k such that F(Al, ... , An-I. 1) i:- O. Put Xl = Xl - AjXn

(1 ~ i ~ n - 1). Show that Xn is integral over the ring A' = k[xl, ... , x~-d, and hence that A is integral over A'. Then apply the inductive hypothesis to A' to complete the proof.

From the proof it follows that Yl, ... , Yr may be chosen to be linear com­binations of Xl, ... , X n • This has the following geometrical interpretation: if k is algebraically closed and X is an affine algebraic variety in kn with coordinate ring A i:- 0, then there exists a linear subspace L of dimension r in kn and a linear mapping of k n onto L which maps X onto L. [Use Exercise 2.]

Nullstellensatz (weak form). 17. Let X be an affine algebraic variety in kn, where k is an algebraically closed field,

and let leX) be the ideal of X in the polynomial ring k[t1o ... , tn ] (Chapter 1, Exercise 27). If 1(X) i:- (1) then X is not empty. [Let A = k[tl,' .• , tn]/I(X) be the coordinate ring of X. Then A i:- 0, hence by Exercise 16 there exists a linear subspace L of dimension ~ 0 in k n and a mapping of X onto L. Hence Xi:- 0.]

70 INTEGRAL DEPENDENCE AND VALUATIONS

Deduce that every maximal ideal in the ring k[t1 , ••• , tn] is of the form (11 - alo.·" tn - an) where aj E k.

18. Let k be a field and let B be a finitely generated k-algebra. Suppose that B is a field. Then B is a finite algebraic extension of k. (This is another version of Hilbert's Nullstellensatz. The following proof is due to Zariski. For other proofs, see (5.24), (7.9).)

Let Xl, ... , Xn generate B as a k-algebra. The proof is by induction on n. If n = 1 the result is clearly true, so assume n > 1. Let A = k[xtl and let K = k(Xl) be the field of fractions of A. By the inductive hypothesis, B is a finite algebraic extension of K, hence each of X2, ••• , Xn satisfies a monic poly­nomial equation with coefficients in K, i.e. coefficients of the form alb where a and b are in A. If/is the product of the denominators of all these coefficients, then each of X2, ••• , X" is integral over A" Hence B and therefore K is integral over A"

Suppose Xl is transcendental over k. Then A is integrally closed, because it is a unique factorization domain. Hence A, is integrally closed (5.12), and there­fore A, = K, which is clearly absurd. Hence Xl is algebraic over k, hence K (and therefore B) is a finite extension of k.

19. Deduce the result of Exercise 17 from Exercise 18.

20. Let A be a subring of an integral domain B such that B is finitely generated over A. Show that there exists s i:- 0 in A and elements Yl, ... , Yn in B, algebraically independent over A and such that B. is integral over B;, where B' = A[Yl, .. " Yn]. [Let S = A - {O} and let K = S -1 A, the field of fractions of A. Then S -1 B is a finitely generated K-algebra and therefore by the normalization lemma (Exercise 16) there exist Xl. .•. ,Xn in S -1 B, algebraically independent over K and such that S -1 B is integral over K[Xl, ... , xn]. Let Zlo"" Zm generate B as an A-algebra. Then each Zj (regarded as an element of S-lB) is integral over K[Xl, . .. , xn]. By writing an equation of integral dependence for each z" show that there exists s E S such that Xj = Yds (l ~ i ~ n) with YI E B, and such that each SZj is integral over B'. Deduce that this s satisfies the conditions stated.]

21. Let A, B be as in Exercise 20. Show that there exists s i:- 0 in A such that, if n is an algebraically closed field and/: A --+ n is a homomorphism for which /(s) i:- 0, then / can be extended to a homomorphism B --+ n. [With the notation of Exercise 20, / can be extended first of all to B' , for example by mapping each YI to 0; then to B: (because /(s) i:- 0), and finally to B. (by Exercise 2, because B. is integral over B;).]

22. Let A, B be as in Exercise 20. If the Jacobson radical of A is zero, then so is the Jacobson radical of B. [Ley v i:- 0 be an element of B. We have to show that there is a maximal ideal of B which does not contain v. By applying Exercise 21 to the ring Bv and its subring A, we obtain an element s i:- 0 in A. Let m be a maximal ideal of A such that s ¢: m, and let k = Aim. Then the canonical mapping A ->- k extends to a homomorphism g of Bv into an algebraic closure n of k. Show that g(v) i:- 0 and that Ker (g) 1\ B is a maximal ideal of B.]

EXERCISES 71

23. Let A be a ring. Show that the following are equivalent: i) Every prime ideal in A is an intersection of maximal ideals.

ii) In every homomorphic image of A the nilradical is equal to the Jacobson radical.

iii) Every prime ideal in A which is not maximal is equal to the intersection of the prime ideals which contain it strictly.

[The only hard part is iii) => ii). Suppose ii) false, then there is a prime ideal which is not an intersection of maximal ideals. Passing to the quotient ring, we may assume that A is an integral domain whose Jacobson radical 91 is not zero. Let/be a non-zero element of 91. Then A, i:- 0, hence A, has a maximal ideal, whose contraction in A is a prime ideal .p such that /¢ %" and which is maximal with respect to this property. Then.p is not maximal and is not equal to the intersection of the prime ideals strictly containing .p.]

A ring A with the three equivalent properties above is called a Jacobson ring.

24. Let A be a Jacobson ring (Exercise 23) and B an A-algebra. Show that if B is either (i) integral over A or (ii) finitely generated as an A-algebra, then B is Jacobson. [Use Exercise 22 for (ii).]

In particular, every finitely generated ring, and every finitely generated algebra over a field, is a Jacobson ring.

25. Let A be a ring. Show that the following are equivalent: i) A is a Jacobson ring; ii) Every finitely generated A-algebra B which is a field is finite over A. [i) => ii). Reduce to the case where A is a subring of B, and use Exercise 21. If SEA is as in Exercise 21, then there exists a maximal ideal m of A not con­taining s, and the homomorphism A -+ Aim = k extends to a homomorphism g of B into the algebraic closure of k. Since B is a field, g is injective, and g(B) is algebraic over k, hence finite algebraic over k. ii) => i). Use criterion iii) of Exercise 23. Let.p be a prime ideal of A which is not maximal, and let B = AI.p. Let/be a non-zerO element of B. Then B, is a finitely generated A-algebra. If it is a field it is finite over B, hence integral over B and therefore B is a field by (5.7). Hence B, is not a field and therefore has a non-zerO prime ideal, whose contraction in B is a non-zero ideal %" such that / ¢ .p'.]

26. Let X be a topological space. A subset of X is locally closed if it is the inter­section of an open set and a closed set, or equivalently if it is open in its closure.

The following conditions on a subset Xo of X are equivalent: (1) Every non-empty locally closed subset of X meets Xo; (2) For every closed set E in X we have E fl Xo = E; (3) The mapping U 1-+ U fl Xo of the collection of open sets of X onto the col­

lection of open sets of Xo is bijective. A subset Xo satisfying these conditions is said to be very dense in X. If A is a ring, show that the following are equivalent:

i) A is a Jacobson ring; ii) The set of maximal ideals of A is very dense in Spec (A);

72 INTEGRAL DEPENDENCE AND VALUATIONS

iii) Every locally closed subset of Spec (A) consisting of a single point is closed. [ii) and iii) are geometrical formulations of conditions ii) and iii) of Exercise 23.]

Valuation rings and valuations

27. Let A, B be two local rings. B is said to dominate A if A is a subring of Band the maximal ideal m of A is contained in the maximal ideal n of B (or, equiva­lently, if m = n n A). Let K be a field and let ~ be the set of all local subrings of K. If ~ is ordered by the relation of domination, show that ~ has maximal elements and that A E ~ is maximal if and only if A is a valuation ring of K. [Use (5.21).]

28. Let A be an integral domain, K its field of fractions. Show that the following are equivalent: (1) A is a valuation ring of K; (2) If a, b are any two ideals of A, then either a s b or b s a.

Deduce that if A is a valuation ring and ~ is a prime ideal of A, then AI' and Aj:p are valuation rings of their fields of fractions.

29. Let A be a valuation ring of a field K. Show that every subring of K which contains A is a local ring of A.

30. Let A be a valuation ring of a field K. The group U of units of A is a subgroup of the multiplicative group K* of K.

Let r = K*jU. If g, 7] E r are represented by x, y E K, define g ~ 7] to mean xy-l EA. Show that this defines a total ordering on r which is compatible with the group structure (Le., g ~ 7] => gw ~ 7]W for all w E r). In other words, r is a totally ordered abelian group. It is called the value group of A.

Let v: K* ->- r be the canonical homomorphism. Show that vex + y) ~ min (v(x), v(y)) for all x, y E K*.

31. Conversely, let r be a totally ordered abelian group (written additively), and let K be a field. A valuation of K with values in r is a mapping v: K* -+ r such that (1) v(xy) = vex) + v(y), (2) vex + y) ~ min (v(x), v(y)), for all x, y E K*. Show that the set of elements x E K* such that vex) ~ 0 is a valuation ring of K. This ring is called the valuation ring of v, and the subgroup v(K*) of r is the value group of v.

Thus the concepts of valuation ring and valuation are essentially equivalent.

32. Let r be a totally ordered abelian group. A subgroup ~ of r is isolated in r if, whenever 0 ~ f3 ~ ex and a E~, we have f3 E~. Let A be a valuation ring of a field K, with value group r (Exercise 31). If:p is a prime ideal of A, show that v(A - :P) is the set of elements ~ 0 in an isolated subgroup ~ of r, and that the ~a~ping so defined of Spec (A) into the set of isolated subgroups of r ~ bi­JectIve.

If:p is a prime ideal of A, what are the value groups of the valuation rings Aj:p, AI'?

33. Let r be -a totally ordered abelian group. We shall show how to construct a field K and a valuation v of K with r as value group. Let k be any field and let

EXERCISES 73

A = k[r] be the group algebra of rover k. By definition, A is freely generated as a k-vector space by elements Xa (a E r) such that XaXp = xa+p. Show that A is an integral domain.

If u = A1Xal + ... + AnXan is any non-zero element of A, where the Ai are all -::j:. 0 and al < ... < an, define vo(u) to be al. Show that the mapping Vo: A - {O} ->- r satisfies conditions (1) and (2) of Exercise 31.

Let K be the field of fractions of A. Show that Vo can be uniquely extended to a valuation v of K, and that the value group of v is precisely r.

34. Let A be a valuation ring and K its field of fractions. Let f: A ->- B be a ring homomorphism such that f*: Spec (B) ->- Spec (A) is a closed mapping. Then if g: B ->- K is any A-algebra homomorphism (i.e., if go f is the embedding of A in K) we have g(B) = A. [Let C = g(B); obviously C ;2 A. Let n be a maximal ideal of C. Since f* is closed, m = n (') A is the maximal ideal of A, whence Am = A. Also the local ring Cn dominates Am. Hence by Exercise 27 we have Cn = A and therefore C sA.]

35. From Exercises 1 and 3 it follows that, if f: A -+ B is integral and C is any A-algebra, then the mapping (f 181 1)*: Spec (B ®A C) ->- Spec (C) is a closed map.

Conversely, suppose that f: A ->- B has this property and that B is an in­tegral domain. Then f is integral. [Replacing A by its image in B, reduce to the case where A s Band f is the injection. Let K be the field of fractions of B and let A' be a valuation ring of K containing A. By (5.22) it is enough to show that A' contains B. By hypothesis Spec (B 0 A A') ->- Spec (A') is a closed map. Apply the result of Exercise 34 to the homomorphism B 0 A A' ->- K defined by b 181 a' f-> ba'. It follows that ba' E A' for all bE B and all a' E A'; taking a' = 1, we have what we want.]

Show that the result just proved remains valid if B is a ring with only finitely many minimal prime ideals (e.g., if B is Noetherian). [Let ~j be the minimal prime ideals. Then each composite homomorphism A -+ B ->- B/b j

is integral, hence A ->- II (B/~I) is integral, hence A ->- B/'iR is integral (where 'iR is the nilradical of B), hence finally A ->- B is integral.]

6

Chain Conditions

So far we have considered quite arbitrary commutative rings (with identity). To go further, however, and obtain deeper theorems we need to impose some finiteness conditions. The most convenient way is in the form of "chain con­ditions". These apply both to rings and modules, and in this chapter we consider the case of modules. Most ofthe arguments are of a rather formal kind and because of this there is a symmetry between the ascending and descending chains-a symmetry which disappears in the caSe of rings as we shall see in subsequent chapters.

Let L be a set partially ordered by a relation ~ (i.e., ~ is reflexive and transitive and is such that x ~ y and y ~ x together imply x = y).

Proposition 6.1. The following conditions on L are equivalent:

i) Every increasing sequence Xl ~ X2 ~. .. in L is stationary (i.e., there exists n such that Xn = Xn + 1 = ... ). ii) Every non-empty subset ofL has a maximal element.

Proof i) => ii). Ifii) is false there is a non-empty subset T ofL with no maximal element, and we can construct inductively a non-terminating strictly increasing sequence in T.

ii) => i). The set (Xm)m:tl has a maximal element, say X n• •

IfL is the set of submodules of a module M, ordered by the relation s;, then i) is called the ascending chain condition (a.c.c. for short) and ii) the maximal condition. A module M satisfying either of these equivalent conditions is said to be Noetherian (after Emmy Noether). If L is ordered by :2, then i) is the descending chain condition (d.c.c. for short) and ii) the minimal condition. A module M satisfying these is said to be Artinian (after Emil Artin).

Examples. 1) A finite abelian group (as Z-module) satisfies both a.c.c. and d.c.c.

2) The ring Z (as Z-module) satisfies a.c.c. but not d.c.c. For if a E Z and a # 0 we have (a) ::> (a2) ::> ••• ::> (an) ::>. •• (strict inclusions).

3) Let G be the subgroup of Q/Z consisting of all elements whose order is a power of p, where p is a fixed prime. Then G has exactly one subgroup Gn of

CHAIN CONDITIONS 75

order pn for each n ;;:: 0, and Go C G1 c· .. C Gn C .•. (strict inclusions) so that G does not satisfy the a.c.c. On the other hand the only proper subgroups of G are the Gn, so that G does satisfy d.c.c.

4) The group H of all rational numbers of the form m/pn (m, nEZ, n ;;:: 0) satisfies neither chain condition. For we have an exact sequence ° -+ Z -+ H-+ G -+ 0, so that H doesn't satisfy d.c.c. because Z doesn't; and H doesn't satisfy a.c.c. because G doesn't.

5) The ring k[x] (k a field, x an indeterminate) satisfies a.c.c. but not d.c.c. on ideals.

6) The polynomial ring k[x1 , X2, ... ] in an infinite number of indeterminates Xn satisfies neither chain condition on ideals: for the sequence (Xl) C (Xl> x2) C . .. is strictly increasing, and the sequence (Xl) ~ (xi) ~ (x~) ~... is strictly decreasing.

7) We shall See later that a ring which satisfies d.c.c. on ideals must also satisfy a.c.c. on ideals. (This is not true in general for modules: see Examples 2,3 above.)

Proposition 6.2. M is a Noetherian A-module =- every submodule of M is finitely generated.

Proof =>: Let N be a submodule of M, and let L be the set of all finitely generated submodules of N. Then L is not empty (since ° E L) and therefore has a maximal element, say No. If No # N, consider the submodule No ,+ Ax where x E N, x ¢ No; this is finitely generated and strictly contains No, so we have a contradiction. Hence N == No and therefore N is finitely generated.

<=:: Let M1 S;; M2 s;; ... be an ascending chain of submodules of M. Then N = U:=l Mn is a submodule of M, hence is finitely generated, say by Xl' ... 'Xr. Say X1EMn\ and let n = maxf_ln,; then each X1EMn, hence Mn = M and therefore the chain is stationary. _

Because of (6.2), Noetherian modules are more important than Artinian modules: the Noetherian condition is just the right finiteness condition to make a lot of theorems work. However, many of the elementary formal properties apply equally to Noetherian and Artinian modules.

Proposition 6.3. Let ° -+ M' ~ M ~ M" -+ ° be an exact sequence of A-modules. Then

i) M is Noetherian =- M' and M" are Noetherian;

ii) Mis Artinian =- M' and M" are Artinian.

Proof We shall prove i); the proof ofii) is similar.

=>: An ascending chain of submodules of M' (or M") gives rise to a chain in M, hence is stationary.

76 CHAIN CONDITIONS

<=: Let (Ln)n;'l be an ascending chain of sub modules of M; then (a- 1(Ln» is a chain in M', and (fJ(Ln) is a chain in M". For large enough n both these chains are stationary, and it follows that the chain (Ln) is stationary. _

Corollary 6.4. If M, (1 ~ i ~ n) are Noetherian (resp. Artinian) A-modules,

so is EBf-l MI' Proof Apply induction and (6.3) to the exact sequence

n n-1

0-+ Mn -+ EB Mi -+ EB M, -+ O. -1= 1 i= 1

A ring A is said to be Noetherian (resp. Artinian) if it is so as an A-module, i.e., if it satisfies a.c.c. (resp. d.c.c.) on ideals.

Examples. 1) Any field is both Artinian and Noetherian; so is the ring Z/{n) (n ¢ 0). The ring Z is Noetherian, but not Artinian (Exercise 2 before (6.2».

2) Any principal ideal domain is Noetherian (by (6.2): every ideal is finitely generated).

3) The ring k[Xh X2 , ... ] is not Noetherian (Exercise 6 above). But it is an integral domain, hence has a field of fractions. Thus a sub ring of a Noetherian ring need not be Noetherian.

4) Let X be a compact infinite Hausdorff space, C(X) the ring of real­valued continuous functions on X. Take a strictly decreasing sequence F1 ::>

F2 ::> ••• of closed sets in X, and let an = {fE C(X):f(Fn) = O}. Then the an form a strictly increasing sequence of ideals in C(X): so C(X) is not a Noetherian ring.

Proposition 6.5. Let A be a Noetherian (resp. Artinian) ring, M a finitely­generated A-module. Then M is Noetherian (resp. Artinian).

Proof M is a quotient of An for some n: apply (6.4) and (6.3). _

Proposition 6.6. Let A be Noetherian (resp. Artinian), a an ideal of A. Then A/a is a Noetherian (resp. Artinian) ring.

Proof By (6.3) A/a is Noetherian (resp. Artinian) as an A-module, hence also as an A/a-module. _

A chain of submodules of a module M is a sequence (M,) (0 ~ i ~ n) of submodules of M such that

M = Mo ::> M1 ::> ..• ::> Mn = 0 (strict inclusions).

The length of the chain is n (the number of "links"). A composition series of M is a maximal chain, that is one in which no extra submodules can be inserted: this is equivalent to saying that each quotient M I - 1 / M, (1 ~ i ~ n) is simple (that is, has no submodules except 0 and itself). -

CHAIN CONDITIONS 77

Proposition 6.7. Suppose that M has a composition series of length n. Then every composition series of M has length n, and every chain in M can be extended to a composition series.

Proof Let I(M) denote the least length of a composition series of a module M. (I(M) = + ex::> if M has no composition series.)

i) N eM=> I(N) < I(M). Let (M,) be a composition series of M of minimum length, and consider the submodules N, = N n M, of N. Since N'_l/N, s Mi-1/M1 and the latter is a simple module, we have either N'_l/N, = M'_l/M" or else M-l = N,; hence, removing repeated terms, we have a com­position series of N, so that I(N) ~ I(M). If I(N) = I(M) = n, then N'_l/N, =

M,-dM, for each i = 1,2,.", n; hence M"'-l = N"'-l' hence M"'-2 = N"'-2" . " and finally M = N.

ii) Any chain in M has length ~ 1(!vI). Let M = Mo ~ Ml ~ ... be a chain of length k. Then by i) we have I(M) > I(Ml) > ... > I(Mk ) = 0, hence I(M) ~ k.

iii) Consider any composition series of M. Ifit has length k, then k ~ I(M) by ii), hence k = I(M) by the definition of I(M). Hence all composition series have the same length. Finally, consider any chain. Ifits length is I(M) it must be a composition series, by ii); ifits length is < I(M) it is not a composition series, hence not maximal, and therefore neW terms can be inserted until the length is I(M) .•

Proposition 6.8. M has a composition series =- M satisfies both chain conditions.

Proof =>: All chains in M are of bounded length, hence both a.c.c. and d.c.c. hold.

¢:: Construct a composition series of M as follows. Since AI = Mo satisfies the maximum condition by (6.1), it has a maximal submodule Ml c Mo. Similarly Ml has a maximal submodule M2 c M 1 , and so on. Thus we have a strictly descending chain Mo ~ Ml ~, " which by d.c.c. must be finite, and hence is a composition series of M. •

A module satisfying both a.c.c. and d.c.c. is therefore called a module of finite length. By (6.7) all composition series of M have the same length I(M), called the length of M. The Jordan-Holder theorem applies to modules offinite length: if (M,)o,;;,,;;,,, and (M;)o,;;,,;;,,, are any two composition series of M, there is a one-to-one correspondence between the set of quotients (M'-l/M')l<l<n and the set of quotients (M;-l/M[)l';;'''"" such that corresponding quotients are isomorphic. The proof is the same as for finite groups.

Proposition 6.9. The length I(M) is an additive function on the class of all A-modules of finite length.

Proof We have to show that ifO -+ M' ~ M ~ M" -+ ° is an exact sequence, then I(M) = I(M') + I(M"). Take the image under IX of any composition

78 CHAIN CONDmONS

series of M' and the inverse image under f3 of any composition series of M"; these fit together to give a composition series of M, hence the result. _

Consider the particular case of modules over a field k, i.e., k-vector spaces:

Proposition 6.10. For k-vector spaces V the following conditions are equiva­lent:

i) finite dimension;

ii) finite length;

iii) a.c.c.;

iv) d.c.c.

Moreover, if these conditions are satisfied, length = dimension.

Proof i) => ii) is elementary; ii) => iii), ii) => iv) from (6.8). Remains to prove iii) => i) and iv) => i). Suppose i) is false, then there exists an infinite sequence (Xn)n>l of linearly independent elements of V. Let Un (resp. Vn) be the vector space spanned by Xl'" ., Xn (resp., Xn+ 1 , Xn+2, ••• ). Then the chain (Un)n;o.l Crespo (Vn)n;o.l) is infinite and strictly ascending (resp. strictly descending).

Corollary 6.11. Let A be a ring in which the zero ideal is a product m1 ... mn of (not necessarily distinct) maximal ideals. Then A is Noetherian if and only if A is Artinian.

Proof Consider the chain of ideals A ~ ml ;;;2 m1 m2 ;;;2 ••• ;;;2 m1··· mn = O. Each factor m l · .. ml-1/m1 ... m l is a vector space over the field A/mi' Hence a.c.c. <=> d.c.c. for each factor. But a.c.c. (resp. d.c.c.) for each factor <=> a.c.c. (resp. d.c.c.) for A, by repeated application of (6.3). Hence a.c.c. <=> d.c.c. for A. _

EXERCISES

1. i) Let M be a Noetherian A-module and u: M -+ M a module homomorphism. If u is surjective, then u is an isomorphism.

ii) If M is Artinian and u is injective, then again u is an isomorphism. [For (i), consider the submodules Ker (un); for (ii), the quotient modules Coker (un).]

2. Let M be an A-module. If every non-empty set of finitely generated sub modules of M has a maximal element, then M is Noetherian.

3. Let M be an A-module and let N1, N2 be sub modules of M. If MINl and MIN2 are Noetherian, so is MI(N1 n N2). Similarly with Artinian in place of Noetherian.

4. Let M be a Noetherian A-module and let a be the annihilator of M in A. Prove that A/a is a Noetherian ring.

If we replace "Noetherian" by "Artinian" in this result, is it still true?

EXERCISES 79

5. A topological space X is said to be Noetherian if the open subsets of X satisfy the ascending chain condition (or, equivalently, the maximal condition). Since closed subsets are complements of open subsets, it comes to the same thing to say that the closed subsets of X satisfy the descending chain condition (or, equivalently, the minimal condition). Show that, if Xis Noetherian, then every subspace of X is Noetherian, and that X is quasi-compact.

6. Prove that the following are equivalent: i) X is Noetherian.

ii) Every open subspace of X is quasi-compact. iii) Every subspace of X is quasi-compact.

7. A Noetherian space is a finite union of irreducible closed subspaces. [Consider the set ~ of closed subsets of X which are not finite unions of irreducible closed subspaces.] Hence the set of irreducible components of a Noetherian space is finite.

8. If A is a Noetherian ring then Spec (A) is a Noetherian topological space. Is the converse true?

9. Deduce from Exercise 8 that the set of minimal prime ideals in a Noetherian ring is finite.

10. If M is a Noetherian module (over an arbitrary ring A) then Supp (M) is a closed Noetherian subspace of Spec (A).

11. Let /: A --- B be a ring homomorphism and suppose that Spec (B) is a Noetherian space (Exercise 5). Prove that f*: Spec (B) --- Spec (A) is a closed mapping if and only if/has the going-up property (Chapter 5, Exercise 10).

12. Let A be a ring such that Spec (A) is a Noetherian space. Show that the set of prime ideals of A satisfies the ascending chain condition. Is the converse true?

7

Noetherian Rings

We recall that a ring A is said to'be Noetherian if it satisfies the following three equivalent conditions:

1) Every non-empty set of ideals in A has a maximal element.

2) Every ascending chain of ideals in A is stationary,

3) Every ideal in A is finitely generated,

(The equivalence of these conditions was proved in (6.1) and (6.2).) Noetherian rings are by far the most important class of rings in commutative

algebra: we have seen some examples already in Chapter 6. In this' chapter we shall first show that Noetherian rings reproduce themselves under various familiar operations-in particular we prove the famous basis theorem of Hilbert. We then proceed to make a number of important deductions from the Noetherian condition, including the existence of primary decompositions.

Proposition 7.1. If A is Noetherian and 4> is a homomorphism of A onto a ring B, then B is Noetherian.

Proof This follows from (6.6), since B ~ A/a, where a = Ker (4)). •

Proposition 7.2. Let A be a subring of B; suppose that A is Noetherian and that B isfinitely generated as an A-module. Then B is Noetherian (as a ring).

Proof By (6.5) B is Noetherian as an A-module, hence also as a B-module. •

Example. B = Z[i], the ring of Gaussian integers. By (7.2) B is Noetherian. More generally, the ring of integers in any algebraic number field is Noetherian.

Proposition 7.3. If A is Noetherian and S is any multiplicatively closed subset of A, then S-IA is Noetherian.

Proof By (3. II-i) and (1.l7-iii) the ideals of S-IA are in one-to-one order­preserving correspondence with the contracted ideals of A, hence satisfy the max­imal condition. (Alternative proof: if a is any ideal of A, then a has a finite set of generators, say Xl> .•• , Xn> and it is clear that S-la is generated by xl/I, ... , xn/l.) •

Corollary 7.4. If A is Noetherian and ~ is a prime ideal of A, then All is Noetherian. •

lIll

NOETHERIAN RINGS 8 I

Theorem 7.5. (Hilbert's Basis Theorem). If A is Noetherian, then the polynomial ring A [x] is Noetherian.

Proof. Let a be an ideal in A[x]. The leading coefficients of the polynomials in a form an ideal I in A. Since A is Noetherian, I is finitely generated, say by a l , ••. , an. For eaGh i = 1, ... , n there is a polynomial It E A[x] of the form It = ajxTj + (lower terms). Let r = max? = 1 rj. The It generate an ideal a' s;; a in A[x].

Letf = axm + (lower terms) be any element of a; we have a E f. If m ;;:: r, write a = 2:f = 1 Uta" where UI E A; then f - 2: Ujltxm

- T, is in a and has degree < m. Proceeding in this way, we can go on subtracting elements of a' from f until we get a polynomial g, say, of degree < r; that is, we have f = g + h, where h E a'.

Let M be the A-module generated by 1, x, ... , x T-

1; then what We have

proved is that a = (a n M) + a'. Now M is a finitely generated A-module, hence is Noetherian by (6.5), hence a n M is finitely generated (as an A-module) by (6.2). If glt ... , gm generate a n M it is clear that the}; and the gj generate a. Hence a is finitely generated and so A[x] is Noetherian. •

Remark. It is also true that A Noetherian => A[[x]] Noetherian (A[[x]] being the ring of formal power series in x with coefficients in A). The proof runs almost parallel to that of (7.5) except that one starts with the terms of lowest degree in the power series belonging to a. See also (10.27).

Corollary 7.6. If A is Noetherian so is A[Xl' ... , x n ].

Proof. By induction on n from (7.5). •

Corollary 7.7. Let B be a finitely-generated A-algebra. If A is Noetherian, then so is B. In particular, every finitely-generated ring, and every finitely generated algebra over a field, is Noetherian.

Proof. B is a homomorphic image of a polynomial ring A[Xl' ... , x n], which is Noetherian by (7.6). •

Proposition 7.B. Let A s;; B s;; .c be rings. Suppose that A is Noetherian, that C is finitely generated as an A-algebra and that C is either· (i) finitely generated as a B-module or (ii) integral over B. Then B is finitely generated as an A -algebra.

Proof. It follows from (5.1) and (5.2) that the conditions (i) and (ii) are equiva­lent in this situation. So we may concentrate on (i).

Let Xl' . : ., Xm generate C as an A-algebra, and let Yl, ... , Yn generate C as a B-module. Then there ~xist expressions of the form

xt = L buYj (blj E B) J

YtYf = L btj"y" (blj" E B). k

(1)

(2)

82 NOETHERIAN RINGS

Let Bo be the algebra generated over A by the b jJ and the blJk' Since A is Noetherian, so is Bo by (7.7), and A S; Bo S; B.

Any element of C is a polynomial in the XI with coefficients in A. Sub­stituting (l) and making repeated Use of (2) shows that each element of C is a linear combination of the y, with coefficients in Bo, and hence C is finitely generated as a Bo-module. Since Bo is Noetherian, and B is a submodule of C, it follows (by (6.5) and (6.2)) that B is finitely generated as a Bo-module. Since Bo is finitely generated as an A-algebra, it follows that B is finitely-generated as an A-algebra. •

Proposition 7.9. Let k be a field, E a finitely generated k-algebra. If E is a field then it is a finite algebraic extension of k.

Proof Let E = k[xl , . .. , xn). If E is not algebraic over k then we can re­number the XI so that Xl' ... , Xr are algebraically independent over k, where r;;: 1, and each of Xr+l>"" Xn is algebraic over the field F = k(xl , ... , xr). Hence E is a finite algebraic extension of F and therefore finitely generated as an F-module. Applying (7.8) to k S; F s; E, it follows that Fis a finitely generated k-algebra, say F = k[Yl, ... , Y.). Each Yj is of the form fjlgj, where It and gj are polynomials in Xl> •.. , X r •

Now there are infinitely many irreducible polynomials in the ring k[xl> ... , x r) (adapt Euclid's proof of the existence of infinitely many prime numbers). Hence there is an irreducible polynomial h which is prime to each of the g, (for example, h = glg2' . 'g. + I would do) and the element h- 1 of F could not be a polynomial in the Yj. This is a contradiction. Hence E is alge­braic over k, and therefore finite algebraic. •

Corollary 7.10. Let k be afield, A afinitely generated k-algebra. Let m be a maximal ideal of A. Then the field Aim is a finite algebraic extension of k. In particular, if k is algebraically closed then Aim ~ k.

Proof Take E = Aim in (7.9). •

(7.1 0) is the so-called "weak" version of Hilbert's Nullstellensatz (= theorem of the zeros). The proof given here is due to Artin and Tate. For its geometrical meaning, and the "strong" form of the theorem, see the Exercises at the end of this chapter.

PRIMARY DECOMPOSITION IN NOETHERIAN RINGS

The next two lemmas show that every ideal -=I (1) in a Noetherian ring has a primary decomposition.

An ideal a is said to be irreducible if

a = b n c => (a = b or a c).

PRIMARY DECOMPOSITION IN NEOTHERIAN RINGS 83

Lemma 7.11. In a Noetherian ring A every ideal is a finite intersection of irreducible ideals.

Proof Suppose not; then the set of ideals in A for which the lemma is false is not empty, hence has a maximal element a. Since a is reducible, We have a = b n c where b ~ a and c ~ a. Hence each of b, c is a finite intersection of irreducible ideals and therefore so is a: contradiction. _

Lemma 7.12. In a Noetherian ring every irreducible ideal is primary.

Proof By passing to the quotient ring, it is enough to show that if the zero ideal is irreducible then it is primary. Let xy = ° with y ¥= 0, and consider the chain of ideals Ann(x) S Ann(x2) s· . '. By the a.c.c., this chain is stationary, i.e., We have Ann(xn) = Ann(xn+l) = ... for some n. It follows that (xn) (I (y) = 0; for if a E (y) then ax = 0, and if a E (xn) then a = bxn, hence bxn+ l = 0, hence b E Ann( xn + I) = Ann( xn), hence bxn = 0; that is, a = 0. Since (0) is irreducible and (y) ¥= ° we must therefore have xn = 0, and this shows that (0) is primary. _

From these two lemmas we have at once

Theorem 7.13. In a Noetherian ring A every ideal has a primary decomposi­tion. _

Hence all the results of Chapter 4 apply to Noetherian rings.

Proposition 7.14. In a Noetherian ring A, every ideal a contains a power of its radical.

Proof Let Xl"", Xk generate rea): say xf' E a (1 ~ i ~ k). Let m = L~= 1 (nl - 1) + 1. Then r(a)m is generated by the products xiI. .. xl'/' with L rl = m; from the definition of m we must have rl ;;:: nl for at least one index i, hence each such monomial lies in a, and therefore r(a)m s a. _

Corollary 7.15. In a Noetherian ring the nilradical is nilpotent. Proof Take a = (0) in (7.14). _

Corollary 7.16. Let A be a Noetherian ring, m a maximal ideal of A, q any ideal of A. Then the following are equivalent:

i) q is m-primary; ii) r(q) = m; iii) mn s q S mfor some n > 0.

Proof i) ,=> ii) is clear; ii) => i) from (4.2); ii) => iii) from (7.14); iii) => ii) by taking radicals: m = r(mn) S r(q) S rem) = m. _

Proposition 7.17. Let a ¥= (1) be an ideal in a Noetherian ring. Then the prime ideals which belong to a are precisely the prime ideals which occur in the set of ideals (a:x) (x E A).

Proof By passing to A/a we may assume that a = 0. Let nf=l ql = ° be a minimal primary decomposition of the zero ideal, and let ~I be the radical of ql'

84 NOETHERIAN RINGS

Let aj = ni"j qj # 0. Then from the proof of (4.5) we have r(Ann(x») = l\ for any x # ° in aj, so that Ann(x) s ~I'

Since qj is ~cprimary, by (7.14) there exists an integer m such that ~r s qj, and therefore al+,r S al n ~r s aj n ql = 0. Let m ~ I be the smallest integer such that al~r = 0, and let x be a non-zero element in aj~r-l. Then l-\X = O,thereforeforsuchanxwehaveAnn(x):2 ~handhenceAnn(x) = Vj'

Conversely, if Ann(x) is a prime ideal V, then r(Ann(x») = V and so by (4.5) ~ is a prime ideal belonging to 0. •

EXERCISES

1. Let A be a non-Noetherian ring and let ~ be the set of ideals in A which are no t finitely generated. Show that ~ has maximal elements and that the maximal elements of:E are prime ideals. [Let a be a maximal element of:E, and suppose that there exist x, YEA such that x rt a and Y rt a and xy E a. Show that there exists a finitely generated ideal Go S a such that Go + (x) = a + (x), and that a = ao + x· (a: x). Since (a:x) strictly contains a, it is finitely generated and therefore so is a.]

Hence a ring in which every prime ideal is finitely generated is Noetherian (I. S. Cohen).

2. Let A be a Noetherian ring and let f = 2:.:'=0 al1x" E A[[x]]. Prove that f is nilpotent if and only if each an is nilpotent.

3. Let a be an irreducible ideal in a ring A. Then the following are equivalent: i) a is primary;

ii) for every multiplicatively closed subset S of A we have (S -la)C = (a :x) for some XE S;

iii) the sequence (a:xn) is stationary, for every x E A.

4. Which of the following rings are Noetherian? i) The ring of rational functions of z having no pole on the circle I zl 1. ii) The ring of power series in z with a positive radius of convergence.

iii) The ring of power series in z with an infinite radius of convergence. iv) The ring of polynomials in z whose first k derivatives vanish at the origin

(k being a fixed integer). v) The ring of polynomials in z, w all of whose partial derivatives with respect

to w vanish for z = O. In all cases the coefficients are complex numbers.

5. Let A be a Noetherian ring, B a finitely generated A-algebra, G a finite group of A-automorphisms of B, and BG the set of all elements of B which are left fixed by every element of G. Show that BG is a finitely generated A-algebra.

6. If a finitely generated ring K is a field, it is a finite field. [If K has characteristic 0, we have Z c: Q s K. Since K is finitely generated over Z it is finitely generated over Q, hence by (7.9) is a finitely generated Q-

EXERCISES 85

module. Now apply (7.8) to obtain a contradiction. Hence K is of characteristic p > 0, hence is finitely generated as a Z/(p)-algebra. Use (7.9) to complete the proof.]

7. Let X be an affine algebraic variety given by a family of equations fa(tl, ... , tn) = 0 (a E I) (Chapter 1, Exercise 27). Show that there exists a finite subset 10 of I such that X is given by the equations fitlo ... , t.,) = 0 for a E 10 •

8. If A[x] is Noetherian, is A necessarily Noetherian?

9. Let A be a ring such that (1) for each maximal ideal m of A, the local ring Am is Noetherian; (2) for each x -::j:. 0 in A, the set of maximal ideals of A which contain x is

finite. Show that A is Noetherian. [Let a -::j:. 0 be an ideal in A. Let mlo ... , mr be the maximal ideals which contain a. Choose Xo -::j:. 0 in a and let mlo ... , mr +s be the maximal ideals which contain Xo. Since mr+lo ... , m r + s do not contain a there exist XJ e a such that Xj rt mr+f (1 ~ j ~ s). Since each Am, (1 ~ i ~ r) is Noetherian, the ex­tension of a in Am, is finitely generated. Hence there exist Xs+ 1, .•. , Xt in a whose images in Am, generate Ant,a for i = I, ... , r. Let ao = (xo, . .. , Xt).

Show that ao and a have the same extension in Am for every maximal ideal m, and deduce by (3.9) that ao = a.]

10. Let M be a Noetherian A-module. Show that M[x] (Chapter 2, Exercise 6) is a Noetherian A[x]-module.

11. Let A be a ring such that each local ring Ap is Noetherian. Is A necessarily Noetherian?

12. Let A be a ring and B a faithfully flat A-algebra (Chapter 3, Exercise 16). If B is Noetherian, show that A is Noetherian. [Use the ascending chain condition.]

13. Let f: A ->- B be a ring homomorphism of finite type and let f*: Spec (B) ->­

Spec (A) be the mapping associated with f Show that the fibers of f* are Noetherian subspaces of B.

Nullstellensatz, strong form

14. Let k be an algebraically closed field, let A denote the polynomial ring k[tl, ., .. , tn ] and let a be an ideal in A. Let Vbe the variety in kn defined by the ideal a, so that V is the set of all x = (Xl. .•. , Xn) E k n such that f(x) = 0 for all fE a. Let I(V) be the ideal of V, i.e. the ideal of all polynomials g E A such that g(x) = 0 for all x E V. Then I( V) = rea). [It is clear that rea) S I( V). Conversely, letf ¢. rea), then there is a prime ideal ~ containing a such thatf¢..)J. Let/be the image of fin B = A/~, let C = B, = B[l/fl, and let m be a maximal ideal of C. Since C is a finitely generated k­algebra we have C/m ~ k, by (7.9). The images XI in C/m of the generators tt of A thus define a point x = (Xlo . .. , X.,) E kn, and the construction shows that x E Vandf(x) i:- 0.]

86 NOETHERIAN RINGS

15. Let A be a Noetherian local ring, m its maximal ideal and k its residue field, and let M be a finitely generated A-module. Then the following are equivalent:

i) Mis free; ii) M is fiat;

iii) the mapping of m ® M into A ® M is injective; iv) Tort (k, M) = O. [To show that iv) => i), let Xl, ... , Xn be elements of M whose images in M/mM form a k-basis of this vector space. By (2.8), the XI generate M. Let F be a free A-module with basis elo ..• , en and define ~: F -+ M by ~(el) = XI' Let E = Ker (cp). Then the exact sequence 0 -+- E -+- F -+- M -+- 0 gives us an exact sequence

O~k ®A E~k ®AF~ k ®AM~O.

Since k ® F and k ® M are vector spaces of the same dimension over k, it follows that 1 ® ~ is an isomorphism, hence k ® E = 0, hence E = 0 by Nakayama's Lemma (E is finitely generated because it is a submodule of F, and A is Noetherian).]

16. Let A be a Noetherian ring, M a finitely generated A-module. Then the following are equivalent:

i) M is a flat A-module; ii) Mop is a free A'P-module, for all prime ideals ~;

iii) Mm is a free Am-module, for all maximal ideals m. In other words, flat = locally free. [Use Exercise 15.]

17. Let A be a ring and M a Noetherian A-module. Show (by imitating the proofs of (7.11) and (7.12») that every submodule N of M has a primary decomposition (Chapter 4, Exercises 20-23).

18. Let A be a Noetherian ring, ~ a prime ideal of A, and M a finitely generated A-module. Show that the following are equivalent:

i) ~ belongs to 0 in M; ii) there exists X E M such that Ann (x) = ~;

iii) there exists a sUbmodule of M isomorphic to A/~. Deduce that there exists a chain of submodules

o = Mo c: M1 c: ••• c M, = M

such that each quotient Md M j -1 is of the form A/~h where ~j is a prime ideal of A.

19. Let a be an ideal in a Noetherian ring A. Let , .

a = n OJ = Cl C, 1_1 ,-

be two minimal decompositions of a as intersections of irreducible ideals. Prove that , = s and that (possible after re-indexing the c,) ,COj) = r(cl) for all i. [Show that for each i = 1, ... " there exists j such that

a = 01 n···n 0 1-1 n CJ n 01+1 n···n Or.]

State and prove an analogous result for modules.

EXERCISES 87

20. Let X be a topological space and let ff be the smallest collection of subsets of X which contains all open subsets of X and is closed with respect to the formation of finite intersections and complements. i) Show that a subset E of X belongs to ff if and only if E is a finite union of sets

of the form U (\ C, where U is open and C is closed. ii) Suppose that X is irreducible and let E E.fF. Show that E is dense in X (i.e., that E = X) if and only if E contains a non-empty open set in X.

21. Let X be a Noetherian topological space (Chapter 6, Exercise 5) and let E S X. Show that E E ff if and only if, for each irreducible closed set Xo S X, either E (\ Xo -::j:. Xo or else E (\ Xo contains a non-empty open subset of Xo. [Suppose E rt ff. Then the collection of closed sets X' S X such that E (\ X' ¢; ff is not empty and therefore has a minimal element Xo. Show that Xo is irreducible and then that each of the alternatives above leads to the conclusion that E n Xo Eff.] The sets belonging to .fF are called the constructible subsets of X.

22. Let X be a Noetherian topological space and let E be a subset of X. Show that E is open in X if and only if, for each irreducible closed subset Xo in X, either E (\ Xo = 0 or else E (\ Xo contains a non-empty open subset of Xo. [The proof is similar to that of Exercise 21.]

23. Let A be a Noetherian ring, I: A -- B a ring homomorphism of finite type (so that B is Noetherian). Let X = Spec (A), Y = Spec (B) and let f*: Y -- X be the mapping associated with I. Then the image under f* of a constructible subset E of Y is a constructible subset of X. [By Exercise 20 it is enough to take E = U (\ C where U is open and C is closed in Y; then. replacing B by a homomorphic image, we reduce to the case where E is open in Y. Since Y is Noetherian, E is quasi-compact and therefore a finite union of open sets of the form Spec (Bg). Hence reduce to the case E = Y. To show that 1·( Y) is constructible, use the criterion of Exercise 21. Let Xo be an irreducible closed subset of X such that f*( Y) (\ Xo is dense in Xo. We have f*(Y) n Xo = f*(f*-l(XO)), andf*-l(Xo) = Spec (AIlJ) 0 A B), where Xo = Spec (AIlJ). Hence reduce to the case where A is an integral domain and/is injec­tive. If Y1 , ••• , Yn are the irreducible components of Y, it is enough to show that some/·( Yi) contains a non-empty open set in X. So finally we are brought down to the situation in which A, B are integral domains and f is injective (and still of finite type); now use Chapter 5, Exercise 21 to complete the proof.]

24. With the notation and hypotheses of Exercise 23, f· is an open mapping -=­I has the going-down property (Chapter 5, Exercise 10). [Suppose I has the going-down property. As in Exercise 23, reduce to proving that E = 1·( Y) is open in X. The going-down property asserts that if lJ E E and lJ' S lJ, then V' E E: in other words, that if Xo is an irreducible closed subset of X and Xo meets E, then E (\ Xo is dense in Xo. By Exercises 20 and 22, E is open in X.l

25. Let A be Noetherian, I: A -+ B of finite type and flat (i.e., B is flat as an A­module). Thenf*: Spec (B) -+ Spec (A) is an open mapping. [Exercise 24 and Chapter 5, Exercise 11.]

4+I.C.A.

88 NOETHERIAN RINGS

Grothendieck groups

26. Let A be a Noetherian ring and let F(A) denote the set of all isomorphism classes of finitely generated A-modules. Let C be the free abelian group generated by F(A). With each short exact sequence 0 ->- M' ->- M ->- MH ->- 0 of finitely generated A-modules we associate the element (M') - (M) + (MH) of C, where (M) is the isomorphism class of M, etc. Let D be the subgroup of C generated by these elements, for all short exact sequences. The quotient group CjD is called the Grothendieck group of A, and is denoted by K(A). If M is a finitely generated A-module, let y(M), or YA(M), denote the image of (M) in K(A).

i) Show that K(A) has the following universal property: for each additive function ,\ on the class of finitely generated A-modules, with values in an abelian group G, there exists a unique homomorphism '\0: K(A) ->- G such that '\(M) = '\o(y(M» for all M.

ii) Show that K(A) is generated by the elements y(A/lJ), where ~1 is a prime ideal of A. [Use Exercise 18.]

iii) If A is a field, or more generally if A is a principal ideal domain, then K(A) ~ z.

iv) Let f: A ->- B be a finite ring homomorphism. Show that restriction of scalars gives rise to a homomorphismk K(B) ->- K(A) such that f{YB(N) = Y A(N) for a B-module N. If g: B ->- C is another finite ring homo­morphism, show that (g 0 f)! = Ji 0 g!.

27. Let A be a Noetherian ring and let Fl(A) be the set of all isomorphism classes of finitely generated flat A-modules. Repeating the construction of Exercise 26 we obtain a group K1(A). Let Yl(M) denote the image of (M) in K1(A).

i) Show that tensor product of modules over A induces a commutative ring structure on K1(A), such that Yl(M)· Yl(N) = Yl(M 181 N). The identity element of this ring is Yl(A).

ii) Show that tensor product induces a K1(A)-module structure on the group K(A), such that Yl(M)'y(N) = y(M 181 N).

iii) If A is a (Noetherian) local ring, then K1(A) ~ z. iv) Let f: A ->- B be a ring homomorphism, B being Noetherian. Show that

extension of scalars gives rise to a ring homomorphism f': K1(A) ->- K1(B) such that P(Yl(M) = Yl(B ®A M). [If M is flat and finitely generated over A, then B ®A M is flat and finitely generated over B.] If g: B ->- Cis another ring homomorphism (with C Noetherian), then (f 0 g)! = f' 0 g!.

v) If f: A -> B is a finite ring homomorphism then

Ji(!!(x)y) = xJi(y)

for x E K1(A), y E K(B). In other words, regarding K(B) as a K1(A)-module by restriction of scalars, the homomorphism f' is a K1(A)-module homo­morphism.

Remark. Since FICA) is a subset of F(A) we have a group homomorphism £: K1(A) ->- K(A), given by £(Yl(M) = y(M). If the ring A is finite-dimensional and regular, Le.,"'if all its local rings A~ are regular (Chapter tl) it can be shown that £ is an isomorphism.

8

Artin Rings

An Artin ring is one which satisfies the d.c.c. (or equivalently the minimal condition) on ideals.

The apparent symmetry with Noetherian rings is however misleading. In fact we will show that an Artin ring is necessarily Noetherian and of a very special kind. In a sense an Artin ring is the simplest kind of ring after a field, and we study them not because of their generality but because of their simplicity.

Proposition 8.1. In an Artin ring A aery prime ideal is maximal.

Proof. Let ~ be a prime ideal of A. Then B = A/~ is an Artinian integral domain. Let x E B, x # 0. By the d.c.c. we have (xn) = (xn+ 1) for some n, hence xn = xn + 1y for some y E B. Since B is an integral domain and x # 0, it follows that we may cancel x n , hence xy = 1. Hence x has an inverse in B, and therefore B is a field, so that ~ is a maximal ideal. _

Corollary 8.2. In an Arlin ring the nilradical is equal to the Jacobson

radical. -

Proposition 8.3. An Artin ring has only a finite number of maximal ideals.

Proof. Consider the set of all finite intersections ntl n· .. n lnn where the ntj

are maximal ideals. This set has a minimal element, say TItl n· .. n ntn ; hence for any maximal ideal nt we have TIt n nt1 n· .. n ntn = nt1 n· .. n mn, and therefore m ;2 nt1 n· .. n TItn . By (1.1t) nt ;2 m i for some i, hence m = ill;

since TIt; is maximal. _

Proposition 8.4. In an Artin ring the nilradical 9, is nilpotent. Proof. By d.c.c. we have 9(k = 9C k +l = ... = a say, for some k > 0. Suppose a # 0, and let 2: denote the set of all ideals 0 such that ao # 0. Then 2: is not empty, since a E 2:. Let c be a minimal element of 2:; then there exists x E c such that xa # 0; we have (x) S c, hence (x) = c by the minimality of c. But (xa)a = xa2 = xa # 0, and xa S (x), hence xa = (x) (again by minimality). Hence x = xy fur some yEa, and therefore x = xy = xy2 = ... = xyn = .... But yEa = 9(k ;2 9(, hence y is nilpotent and therefore x = xyn = 0. This contradicts the choice of x, therefore a = 0. _

By a chain of prime ideals of a ring A we mean a finite strictly increasing sequence Po c ~ 1 c ... C ~n; the length of the chain is n. We define the

89

90 ARTlN RINGS

dimension of A to be the supremum of the lengths of all chains of prime ideals in A: it is an integer;;:: 0, or + 00 (assuming A # 0). A field has dimension 0; the ring Z has dimension 1.

Theorem 8.5. A ring A is Artin <0> A is Noetherian and dim A = O. Proof =>: By (8.1) We have dim A = O. Let TIl, (1 ~ i ~ n) be the distinct maximal ideals of A (8.3). Then Df=l nt~ S (nf=l mi)k = ink = O. Hence by (6.11) A is Noetherian.

¢:: Since the zero ideal has a primary decomposition (7.13), A has only a finite number of minimal prime ideals, and these are all maximal since dim A = O. Hence in = nf=l m, say; We have ink = 0 by (7.15), hence Df=l m~ = 0 as in the previous part of the proof. Hence by (6.11) A is an Artin ring. •

If A is an Artin local ring with maximal ideal m, then nt is the only prime ideal of A and therefore m is the nilradical of A. Hence every element of m is nilpotent, and m itself is nilpotent. Every element of A is either a unit or is nilpotent. An example of such a ring is Z/(pn), where p is prime and n ;;:: 1.

Proposition 8.6. Let A be a Noetherian local ring, m its maximal ideal. Then exactly one of the following two statements is true: i) mn # mHl for all n;

ii) mn = 0 for some n, in which case A is an Arlin local ring.

Proof Suppose mn = mn + 1 for some n. By Nakayama's lemma (2.6) we have mn = O. Let ~ be any prime ideal of A. Then mn s ~, hence (taking radicals) nt = ~. Hence nt is the only prime ideal of A and therefore A is Artinian. •

Theorem 8.7. (structure theorem for Artin rings). An Artin ring A is uniquely (up to isomorphism) a finite direct product of Artin local rings.

Proof Let m i (1 ~ i ~ n) be the distinct maximal ideals of A. From the proof of (8.5) we have rlf=l m~ = 0 for some k > O. By (1.16) the ideals m~ are coprime in pairs, 'hence n m~ = D m~ by (1.10). Consequently by (1.10) again the natural mapping A -+ Df= 1 (A/m~) is an isomorphism. Each A/nt~ is an Artin local ring, hence A is a direct product of Artin local rings.

Conversely, suppose A ~ D7'=l Ai> where the Ai are Artin local rings. Then for each i we have a natural surjective homomorphism (projection on the ith factor) 4>i: A -+ A,. Let ai = Ker (4),). By (1.10) the a, are pairwise coprime, and n a, = O. Let qi be the unique prime ideal of Ai> and let ~i be its contraction 4>,-l(q,). The ideal ~, is prime and therefore maximal by (8.1). Since q, is nil­potent it follows that a, is ~i-primary, and hence n ai = (0) is a primary de­composition of the zero ideal in A. Since the a, are pairwise coprime, so are the ~b and they are therefore isolated prime ideals of (0). Hence all the primary components ai are isolated, and therefore uniquely determined by A, by the 2nd uniqueness theorem (4.11). Hence the rings AI ~ A/al are uniquely determined by A .•

EXERCISES 91

.. Example. A ring with only one prime ideal need not be Noetherian (and hence not an Artin ring). Let A = k[xl> x 2 , ••• ] be the polynomial ring in a countably infinite set of indeterminates Xn over a field k, and let a be the ideal (Xl> x~, ... , x:, .. . ). The ring B = A/a has only one prime ideal (namely the image of (Xl' x2 , ••• , xn, ..• )), hence B is a local ring of dimension O. But B is not Noetherian, for it is not difficult to see that its prime ideal is not finitely gen­erated.

If A is a local ring, m its maximal ideal, k = A/m its residue field, the A-module m/m2 is annihilated by m and therefore has the structure of a k-vector space. If m is finitely generated (e.g., if A is Noetherian), the images in m/m2 of a set of generators of m will span mjm2 as a vector space, and therefore dimk (m/m2) is finite. (See (2.8).)

Proposition 8.8. Let A be an Artin local ring. Then the following are equivalent,'

i) every ideal in A is principal;

ii) the maximal ideal m is principal;

iii) dimk (m/m2) ~ 1.

Proof i) => ii) => iii) is clear. iii) => i): Ifdimk (m/m2) = 0, then m = m2, hence m = 0 by Nakayama's

lemma (2.6), and therefore A is a field and there is nothing to prove. If dimk (m/m2) = 1, then m is a principal ideal by (2.8) (take M = m

there), say m = (x). Let a be an ideal of A, other than (O) or (1). We have m = 91, hence m is nilpotent by (8.4) and therefore there exists an integer r such that as;;; mT, a $ mT+l; hence there exists yEa such that y = axT, Y ti (xT + 1); consequently a ti (x) and a is a unit in A. Hence x T E a, therefore mT = {xT

} s;;; a and hence a = mT = (xT). Hence a is principal. _

Example. The rings Zj{pn) (p prime), k[x]f{r) (f irreducible) satisfy the conditions of (8.7). On the other hand, the Artin local ring k[x2, X3]/{X4) does not: here m is generated by x2 and x3 (mod X4), so that m2 = 0 and dim (m/m2) = 2. ..

EXERCISES

1. Let ql n· .. (") qn = 0 be a minimal primary decomposition of the zero ideal in a Noetherian ring, and let qj be ~j-primary. Let ~\T) be the rth symbolic power of ~I (Chapter 4, Exercise 13). Show that for each i = 1, ... , n there exists an integer rl such that ~iTI) s ql'

4*

, Suppose qj is an isolated primary component. Then ApI is an Artin local ring, hence ifmj is its maximal ideal we have mj = 0 for all sufficiently large r, hence qj = ~l') for all large r.

92 ARTIN RINGS

If ql is an embedded primary component, then Ap. is not Artinian, hence the powers mr are all distinct, and so the pi') are all distinct. Hence in the given primary decomposition we can replace ql by any of the infinite set of Pi-primary ideals pi') where r ;;:. rio and so there are infinitely many minimal primary decompositions of 0 which differ only in the PI-component.

2. Let A be a Noetherian ring. Prove that the following are equivalent: i) A is Artinian;

ii) Spec (A) is discrete and finite; iii) Spec (A) is discrete.

3. Let k be a field and A a finitely generated k-algebra. Prove that the following are equivalent: i) A is Artinian;

ii) A is a finite k-algebra. [To prove that i) ~ ii), use (8.7) to reduce to the case where A is an Artin local ring. By the Nullstellensatz, the residue field of A is a finite extension of k. Now use the fact that A is of finite length as an A-module. To prove ii) => i), observe that the ideals of A are k-vector subspaces and therefore satisfy d.c.c.)

4. Let f: A --;;. B be a ring homomorphism of finite type. Consider the following statements:

i) f is finite; ii) the fibres of f* are discrete subspaces of Spec (B);

iii) for each prime ideal P of A, the ring B ~A k(p) is a finite k(p)-algebra (k(p) is the residue field of All);

iv) the fibres of f* are finite. Prove that i) ~ ii) ¢> iii) => iv). [Use Exercises 2 and 3.)

If f is integral and the fibres of f* are finite, is f necessarily finite?

5. In Chapter 5, Exercise 16, show that X is a finite covering of L (Le., the number of points of X lying over a given point of L is finite and bounded).

6. Let A be a Noetherian ring and q a p-primary ideal in A. Consider chains of primary ideals from q to p. Show that all such chains are of finite bounded length, and that all maximal chains have the same length.

9

Discrete Valuation Rings and

Dedekind Domains

As we have indicated before, algebraic number theory is one of the historical sources of commutative algebra. In this chapter we specialize down to the case of interest in number theory, namely to Dedekind domains. We deduce the unique factorization of ideals in Dedekind domains from the general primary decomposition theorems. Although a direct approach is of course possible one obtains more insight our way into the precise context of number theory in commutative algebra. Another important class of Dedekind domains occurs in connection with non-singular algebraic curves. In fact the geometrical picture of the Dedekind condition is: non-singular of dimension one.

The last chapter dealt with Noetherian rings of dimension O. Here we start by considering the next simplest case, namely Noetherian integral domains of dimension one: i.e., Noetherian domains in which every non-zero prime ideal is maximal. The first result is that in such a ring we have a unique factorization theorem for ideals:

Proposition 9.1. Let A be a Noetherian domain of dimension 1. Then every non-zero ideal a in A can be uniquely expressed as a product of primary ideals whose radicals are all distinct.

Proof Since A is Noetherian, a has a minimal primary decomposition a = nf= 1 ql by (7.13), where each ql is say ;>I-primary. Since dim A = 1 and A is an integral domain, each non-zero prime ideal of A is maximal, hence the ;>1 are distinct maximal ideals (since PI ;2 ql ;2 a =p 0), and are therefore pairwise coprime. Hence by (1.16) the ql are pairwise coprime and therefore by (1.10) we have f1 qj = n ql' Hence a = f1 ql'

Conversely, if a = f1 qh the same argument shows that a = n ql; this is a minimal primary decomposition of a, in which each ql is an isolated primary component, and is therefore unique by (4.11). •

Let A be a Noetherian domain of dimension one in which every primary ideal is a prime power. By (9.1), in such a ring we shall have unique factorization of non-zero ideals into products of prime ideals. If we localize A with respect to a non-zero prime ideal p we get a local ring All satisfying the same conditions

93

94 DISCRETE VALUATION RINGS AND DEDEKIND DOMAINS

as A, and therefore in All every non-zero ideal is a power of the maximal ideal. Such local rings can be characterized in other ways.

DISCRETE VALUATION RINGS

Let Kbe a field. A discrete valuation on Kis a mapping v of K* onto Z (where K* = K - {O} is the multiplicative group of K) such that

1) v(xy) = vex) + v(y), i.e., v is a homomorphism;

2) vex + y) ~ min(v(x),v(y»).

The set consisting of 0 and all x E K* such that vex) ~ 0 is a ring, called the valuation ring of v. It is a valuation ring of the field K. It is sometimes convenient to extend v to the whole of K by putting v(O) = + 00.

Examples. The two standard examples are:

1) K = Q. Take a fixed prime p, then any non zero x E Q can be written uniquely in the form pay, where a E Z and both numerator and denominator of y are prime to p. Define vp(x) to be a. The valuation ring of Vp is the local ring Z(p).

2) K = k(x), where k is a field and x an indeterminate. Take a fixed irreducible polynomialfE k[x] and define Vf just as in 1). The valuation ring of Vf is then the local ring of k[x] with respect to the prime ideal (f).

An integral domain A is a discrete valuation ring if there is a discrete valua­tion v of its field of fractions K such that A is the valuation ring of v. By (5.18), A is a local ring, an6 its maximal ideal m is the set of all x E K such that vex) > O.

If two elements x, y of A have the same value, that is if vex) = v(y), then V(xy-l) = 0 and therefore u = xy-l is a unit in A. Hence (x) = (y).

If a of 0 is an ideal in A, there is a least integer k such that vex) = k for some x E a. It follows that a contains every YEA with v(y) ~ k, and therefore the only ideals of 0 in A are the ideals m" = {y E A: v(y) ~ k}. These form a single chain m ::> m2 ::> m3 ::> •• " and therefore A is Noetherian.

Moreover, since v: K* --+ Z is surjective, there exists x Em such that vex) = 1, and then m = (x), and m" = (x") (k ~ 1). Hence m is the only non-zero prime ideal of A, and A is thus a Noetherian local domain of dimension one in which every non-zero ideal is a power of the maximal ideal. In fact many of these properties are characteristic of discrete valuation rings.

Proposition 9.2. Let A be a Noetherian local domain of dimension one, m its maximal ideal, k = Aim its residue field. Then the following are equivalent,'

i) A is a discrete valuation ring;

ii) A is integrally closed;

iii) m is a principal ideal;

DEDEKIND DOMAINS 95

iv) dimk (m/m2) = 1; v) Every non-zerO ideal is a power ofm;

vi) There exists x E A such that every non-zero ideal is of the form (xk), k ~ O.

Proof Before we start going the rounds, we make two remarks:

(A) If a is an ideal of 0, (1), then a is m-primary and a ;2 mil for some n. For rea) = m, since m is the only non-zero prime ideal; now use (7.16).

(8) m" of m" + 1 for all n ~ O. This follows from (8.6). i) => ii) by (5.18).

ii) => iii). Let a E m and a of O. By remark (A) there exists an integer n such that m" £: (a), mn

-1 $ (a). Choose bE m"- 1 and b ¢: (a), and let x =

alb E K, the field of fractions of A. We have x- 1 rf= A (since b rf= (a»), hence x- 1 is not integral over A, and therefore by (5.1) we have x- 1m $ m (forif x- 1m £: m, m would be a faithful A[x- 1 ]-module, finitely generated as an A-module). But x- 1m £: A by construction of x, hence x- 1m = A and therefore m = Ax = (x).

iii) => iv). By (2.8) we have dimk (m/m2) ~ 1, and by remark (B) m/m2 of O.

iv) => v). Let a be an ideal of (0), (1). By remark (A) we have a 2 m" for some n; from (8.8) (applied to Aim") it follows that a is a power of m. .

v) => vi). 8y remark (B), m of m 2, hence there exists x E m, x rf= m2. But

(x) = mT by hypothesis, hence r = 1, (x) = m, (xk) = mk.

vi) => i). Clearly (x) = m, hence (xk) of (Xk + 1) by remark (B). Hence if a is any non-zero element of A, we have (a) = (Xk) for exactly one value of k. Define L'(a) = k and extend v to K* by defining v(ab -1) = v(a) - v(b). Check that v is well-defined and is a discrete valuation, and that A is the valuation ring ofv .•

DEDEKIND DOMAINS

Theorem 9.3. Let A be a Noetherian domain of dimension one. Then the following are equivalent:

i) A is integrally closed; ii) Every primary ideal in A is a prime power;

iii) Every local ring All (.):> of 0) is a discrete valuation ring.

Proof i) -¢> iii) by (9.2) and (5.13). ii) -¢> iii). Use (9.2) and the fact that primary ideals and powers of ideals

behave well under localization: (4.8), (3.11). •

A ring satisfying the conditions of (9.3) is called a Dedekind domain.

Corollary 9.4. In a Dedekind domain every non-zero ideal has a unique factorization as a product of prime ideals.

Proof (9.1) and (9.3). •

96 DISCRETE VALUATION RINGS AND DEDEKIND DOMAINS

Examples. 1) Any principal ideal domain A. For A is Noetherian (since every ideal is finitely generated) and of dimension one (Example 3 after (1.6)). Also every local ring Ap(p "# 0) is a principal ideal domain, hence by (9.2) a discrete valuation ring; hence A is a Dedekind domain by (9.3).

2) Let K be an algebraic number field (a finite algebraic extension of Q). Its ring of integers A is the integral closure of Z in K. (For example, if K = Q(i), then A = Z[i], the ring of Gaussian integers.) Then A is a Dedekind domain:

Theorem 9.5. The ring of integers in an algebraic number field K is a Dedekind domain.

Proof K is a separable extension of Q (because the characteristic is zero), hence by (5.17) there is a basis VI, ... , V" of Kover Q such that A S;; L Zl').

Hence A is finitely generated as a Z-module and therefore Noetherian. Also A is integrally closed by (5.5). To complete the proof we must show that every non­zero prime ideal p of A is maximal, and this follows from (5.8) and (5.9): (5.9) shows that p ("\ Z "# 0, hence p ("\ Z is a maximal ideal of Z and therefore p is maximal in A by (5.8). •

Remark. The unique factorization theorem (9.4) was originally proved for rings of integers in algebraic number fields. The uniqueness theorems of Chapter 4 may be regarded as generalizations of this result: prime powers have to be replaced by primary ideals, and products by intersections.

FRACTIONAL IDEALS

Let A be an integral domain, K its field of fractions. An A-submodule M of K is a fractional ideal of A if xM s;; A for some x "# ° in A. In particular, the "ordinary" ideals (now called integral ideals) are fractional ideals (take x = 1). Any element U E K generates a fractional ideal, denoted by (u) or Au, and called principal. If M is a fractional ideal, the set of all x E K such that x M s;; A is de­noted by (A: M).

Every finitely generated A-submodule M of K is a fractional ideal. For if M is generated by Xl> ... , x" E K, we can write Xl = YiZ-I (1 ~ i ~ n) where )'i

and z are in A, and then zM S;; A. Conversely, if A is Noetherian, every fractional ideal is finitely generated, for it is of the form x -In for some integral ideal a.

An A-submodule M of K is an invertible ideal if there exists a submodule N of K such that MN = A. The module N is then unique and equal to (A: M), for we have N s;; (A: M) = (A: M)MN S;; AN = N. It follows that M is finitely generated, and therefore a fractional ideal: for since M· (A: M) = A there exist XI E M and )'t E (A: M) (1 ~ i ~ n) such that L XtYt = 1, and hence for any x E M we have x = L (YtX)xl: each )'iX E A, so that M is generated by

Xl>"" x". Clearly every non-zero principal fractional ideal (u) is invertible, its inverse

being (u- I ). The invertible ideals form a group with respect to multiplication, whose identity element is A = (1).

FRACTIONAL IDEALS 97

Invertibility is a local property:

Proposition 9.6. For a/ractional ideal M, the following are equivalent:

i) M is invertible;

ii) M is finitely generated and, for each prime ideal p, Mp is invertible:

iii) M is finitely generated and, for each maximal ideal m, Mm is invertible.

Proof i) =:> ii): Ap = (M·(A:M»)p = Mp'(Ap:Mp) by (3.11) and (3.15) (for M is finitely generated, because invertible).

ii) =:> iii) as usual.

iii =:> i): Let a = M·(A:M), which is an integral ideal. For each maximal ideal m we have am = Mm·(Am:Mm) (by (3.11) and (3.15» = Am because Mm is invertible. Hence a $ m. Consequently a = A and therefore M is invertible. _

Proposition 9.7. Let A be a local domain. Then A is a discrete valuation ring -¢> every non-zero fractional ideal of A is invertible.

Proof =:>. Let x be a generator of the maximal ideal m of A, and let M of 0 be a fractional ideal. Then there exists YEA such that yM sA: thus y M is an integral ideal, say (xT

), and therefore M = (xT-

S) where s = v(y).

<=: Every non-zero integral ideal is invertible and therefore finitely gen­erated,' so that A is Noetherian. It is therefore enough to prove that every non­zero integral ideal is a power of m. Suppose this is false; let ~ be the set of non­zero ideals which are not powers of m, and let a be a maximal element of~. Then a of m, hence a em; hence m -la em-1m = A is a proper (integral) ideal, and m -la ;2 a. If m -la = a, then a = ma and therefore a = 0 by Nakayama's lemma (2.6); hence m- la :::> a and hence m-1a is a power of m (by the maximality of a). Hence a is a power of m: contradiction. _

The "global" counterpart of (9.7) is

Theorem 9.8. Let A be an integral domain. Then A is a Dedekind domain -¢>

every non-zero fractional ideal of A is invertible.

Proof =:>: Let M of 0 be a fractional ideal. Since A is Noetherian, M is finitely generated. For each prime ideal p of 0, Mp is a fractional ideal of 0 of the discrete valuation ring A p, hence is invertible by (9.7). Hence M is invert~ble, by (9.6).

<=: Every non-zero integral ideal is invertible, hence finitely generated, hence A is Noetherian. We shall show that each Ap (p "# 0) is a discrete valua­tion ring. For this it is enough to show that each integral ideal of 0 in Ap is invertible, and then use (9.7). Let b "# 0 be an (integral) ideal in A p, and let a = be = b n A. Then a is invertible, hence b = ap is invertible by (9.7). •

98 DISCRETE VALUATION RINGS AND DEDEKIND DOMAINS

Corollary 9.9. If A is a Dedekind domain, the non-zero fractional ideals of A form a group with respect to multiplication. _

This group is called the group of ideals of A; we denote it by I. In this terminology (9.4) says that I is a free (abelian) group, generated by the non-zero prime ideals of A.

Let K* denote the multiplicative group of the field of fractions K of A. Each u E K* defines a fractional ideal (u), and the mapping u f---7 (u) is a homo­morphism <p: K* -+ l. The image P of <p is the group of principal fractional ideals: the quotient group H = IjP is called the ideal class group of A. The kernel U of <p is the set of all u E K* such that (u) = (I), so that it is the group of units of A. We have an exact sequence

1 -+ U -+ K* -+ I -+ H -+ 1.

Remark. For the Dedekind domains that arise in number theory, there are classical theorems relating to the groups Hand U. Let K be an algebraic number field and let A be its ring of integers, which is a Dedekind domain by (9.5). In this case:

1) H is a finite group.· Its order h is the class number of the field K. The fol­lowing are equivalent: (i) h = 1; (ii) I = P; (iii) A is a principal ideal domain; (iv) A is a unique factorization domain.

2) U is a finitely-generated abelian group. More precisely, we can specify the number of generators of U. First, the elements of finite order in U are just the roots of unity which lie in K, and they form a finite cyclic group W; UjW is torsion-free. The number of generators of UjWis given as follows: if(K:Q) = n there are n distinct embeddings K -+ C (the field of complex numbers). Of these, say r1 map K into R, and the rest pair off (if ex is one, then w 0 ex is another, where w is the automorphism of C defined by w(z) = z) into say r2 pairs: thus r1 + 2r2 = n. The number of generators of Uj W is then r1 + r2 - 1.

The proofs of these results belong to algebraic number theory and not to commutative algebra: they require techniques of a different nature from those used in this book.

Examples. 1) K = Q(-v=t); n = 2, r1 = 0, r2 = 1, r1 + r2 - 1 = 0. The only units in Z(i] = A are the four roots of unity ± 1, ± i.

2) K = Q(V2); n = 2, r1 = 2, r2 = 0, r1 + r2 - 1 = 1. W = {±I}, and UjW is infinite cyclic. In fact the units in A = Z[V2] are ± (1 + V2?, where n is any rational integer.

EXERCISES 99

EXERCISES

1. Let A be a Dedekind domain, S a multiplicatively closed subset of A. Show that S -lA is either a Dedekind domain or the field of fractions of A.

Suppose that S :;C A - {O}, and let H, H' be the ideal class groups of A and S -1 A respectively. Show that extension of ideals induces a surjective homo­morphism H ~ H'.

2. Let A be a Dedekind domain. If f = ao + alx + ... + anxn is a polynomial with coefficients in A, the content of f is the ideal c(f) = (ao, ... , an) in A. Prove Gauss's lemma that c(fg) = c(f)c(g). [Localize at each maximal ideal.]

3. A valuation ring (other than a field) is Noetherian if and only if it is a discrete valuation ring.

4. Let A be a local domain which is not a field and in which the maximal ideal m is principal and n:=l mn = O. Prove that A is a discrete valuation ring.

5. Let M be a finitely-generated module over a Dedekind domain. Prove that Mis flat =- M is torsion-free. (Use Chapter 3, Exercise 13 and Chapter 7, Exercise 16.]

6. Let M be a finitely-generated torsion module (T(M) = M) over a Dedekind dorriain A. Prove that M is uniquely representable as a finite direct sum of mod­ules A/pi", where p, are non-zero prime ideals of A. [For each p i= 0, Mll is a torsion All-module; use the structure theorem for modules over a principal ideal domain.]

7. Let A be a Dedekind domain and a :;c 0 an ideal in A. Show that every ideal in A/a is principal.

Deduce that every ideal in A can be generated by at most 2 elements.

8. Let a, Ii, c be three ideals in a Dedekind domain. Prove that

a (\ (Ii + c) = (a (\ Ii) + (a (\ c)

a + (Ii (\ c) = (a + Ii) (\ (a + c). (Localize. ]

9. (Chinese Remainder Theorem). Let at. ... , an be ideals and let Xl,.'" Xn be elements in a Dedekind domain A. Then the system of congruences x == X, (mod a,) (1 ~ i ~ n) has a solution x in A =- X, == Xj (mod a, + aj) when­ever i :;C j.

[This is equivalent to saying that the sequence of A-modules

is exact, where 4> and .jJ are defined as follows: 4>(x) = (x + al, . .. , X + an); .jJ(Xl + al, ... , Xn + an) has (i, j)-component x, - Xj + a, + aj. To show that this sequence is exact it is enough to show that it is exact when localized at any p >F 0: in other words we may assume that A is a discrete valuation ring, and then it is easy.]

S+I.C.A.

10

Completions

In classical algebraic geometry (i.e. over the field of complex numbers) we can use transcendental methods. This means that we regard a rational function as an analytic function (of one or more complex variables) and consider its power series expansion about a point. In aQstract algebraic geometry the best we can do is to consider the corresponding formal power series. This is not so powerful as in the holomorphic case but it can be a very useful tool. The process of replacing polynomials by formal power series is an example of a general device known as completion. Another important instance of completion occurs in number theory in the formation of p-adic numbers. If p is a prime number in Z we can work in the various quotient rings Zjp"Z: in other words, we can try and solve congruences modulo pn for higher and higher values of n. This is analogous to the successive approximations given by the terms of a Taylor expansion and, just as it is convenient to introduce formal power series, so it is convenient to introduce the p-adic numbers, these being the limit in a certain sense of Zjp"Z as n -+ 00. In one respect, however, the p-adic numbers are more complicated than formal power series (in, say, one variable x). Whereas the polynomials of degree n are naturally embedded in the power series, the group Zjp"Z cannot be embedded in Z. Although a p-adic integer can be thought of as a power series L a"p" (0 :s;; an < p) this representation does not behave well under the ring operations.

In this chapter we shall describe the general process of "adic" completion­the prime p being replaced by a general ideal. It is most conveniently expressed in topological terms but the reader should beware of using the topology of.the real numbers as an intuitive guide. Instead he should think of the power series topology in which a power series is "small" if it has only terms of high order. Alternatively he can think of the p-adic topology on Z, in which an integer is "small" if it is divisible by a high power of p.

Completion, like localization, is a method of simplifying things by con­centrating attention near a point (or prime). It is, however, a more drastic simplification than localization. For example, in algebraic geometry the local ring of a non-singular point on a variety of dimension n always has for its completion the ring offormal power series in n variables (this will essentially be proved in Chapter 11). On the other hand the local rings of two such points cannot be isomorphic unless the varieties on which they lie are birationally

1M

TOPOLOGIES AND COMPLETIONS 101

equivalent (this means that the fields of fractions of the two local rings are isomorphic).

Two of the important properties of localization are that it preserves exact­ness and the Noetherian property. The same is true for completion-when we restrict to finitely-generated modules-but the proofs are much harder and take up most of this chapter. Another important result is the theorem of Krull which identifies the part of a ring which is "killed" by completion. Roughly speaking, Krull's Theorem is the analogue of the fact that an analytic function is determined by the coefficients of its Taylor expansion. This analogy is clearest for a Noetherian local ring in which case the theorem just asserts that n m" = 0 where m is the maximal ideal. Both Krull's Theorem and the exactness of com­pletion are easy consequences of the well-known "Artin-Rees Lemma", and we accord this lemma a central place in our treatment.

For the study of completions we shall find it necessary to introduce graded rings. The prototype of a graded ring is the ring of polynomials k[Xl' ... , x,,), the grading being the usual one obtained by taking the degree of each variable to be 1. Just as ungraded rings are the foundation for. affine algebraic geometry, so graded rings are the foundation for projective algebraic geometry. They are therefore of considerable geometric importance. The important construction of the associated graded ring Ga(A) of an ideal a of A, which we shall meet, has a very definite geometrical interpretation. For example, if A is the local ring of a point P on a variety V with a as maximal ideal, then Ga(A) corresponds to the projective tangent cone at P, i.e. all the lines through P which are tangent to V at P. This geometrical picture should help to explain the significance of Ga(A) in connection with the properties of V near P and in particular in connection with the study of the completion A.

TOPOLOGIES AND COMPLETIONS

Let G be a topological abelian group (written additively), not necessarily Hausdorff: thus G ~s both a topological space and an abelian group, and the two structures on G are compatible in the sense that the mappings G x G -+ G and G -+ G, defined by (x, y) 1-+ X + Y and x 1-+ - x respectively, are continuous. If {O} is closed in G, then the diagonal is closed in G x G (being the inverse image of {O} under the mapping (x,y) 1-+ x-y) and so G is Hausdorff. If a is a fixed element of G the translation Ta defined by TaCx) = x + a is a homeo­morphism of G onto G (for Ta is continuous, and its inverse is T _ ,,); hence if U is any neighborhood of 0 in G, then U + a is a neighborhood of a in G, and conversely every neighborhood of a appears in this form. Thus the topology of G is uniquely determined by the neighborhoods of 0 in G.

Lemma 10.1. Let H be the intersection of all neighborhoods of 0 in G. Then i) H is a subgroup.

102 COMPLETIONS

ii) H is the closure of {O}.

iii) G/H is Hausdorff.

iv) G is Hausdorff -¢> H = O. Proof i) follows from the continuity of the group operations. For ii) we have:

x E H -¢> 0 E X - U for all neighborhoods U of 0 -¢> x E {O}.

ii) implies that the co sets of H are all closed; thus points are closed in Gj H and so G/H is Hausdorff. Thus H = 0 => G is Hausdorff, and the converse is trivial. _

Assume for simplicity that 0 E G has a countable fundamental system of neighborhoods. Then the completion G of G may be defined in the usual way by means of Cauchy sequences. A Cauchy sequence in G is defined to be a sequence (xv) of elements of G such that, for any neighborhood U of 0, there exists an integer s(U) with the property that

xl' - Xv E U for allp-, v ~ s(U).

Two Cauchy sequences are equivalent if Xv - Yv -r 0 in G. The set of all equivalence classes of Cauchy sequences is denoted by G. If (xv), (Yv) are Cauchy sequences, so is (xv + Yv), and its class in G depends only on the classes of (xv) and (Yv). Hence we have an addition in G with respect to which G is an abelian group. For each x E G the class of the constant sequence (x) is an element </>(x) of G, and </>: G -r G is a homomorphism of abelian groups. Note that </> is not in general injective. In fact we have

Ker</> = n U

where U runs through all neighborhoods of 0 in G, and so by (10.1) </> is injective if and only if G is Hausdorff.

If H is another abelian topological group and f: G -r H a continuous homomorphism, then the image under f of a Cauchy sequence in G is a Cauchy sequence in H, and therefore f induces a homomorphistn /: G -r fI, which is

/"-... continuous. If we have G !.-r H!...r K, then g 0 f = go J

So far we have been quite general and G could for instance have been the additive group of rationals with the usual topology, so that G would be the real numbers. Now, however, we restrict ourselves to the special kind of topologies occurring in commutative algebra, namely we assume that 0 E G has a funda­mental system of neighborhoods consisting of subgroups. Thus we have a sequence of subgroups

G = Go ;2 G1 ;2 ... ;2 G" ;2 ...

and U s; G is a neighborhood of 0 if and only if it contains some G". A typical example is the p-adic topology on Z, in which G" = p"Z. Note that in such topologies the subgroups G" of G are both open and closed. In fact if g E G"

TOPOLOGIES AND COMPLETIONS 103

then g + G" is a neighborhood of g; since g + G" S;; G" this shows G" is open. Hence for any h the coset h + G" is open and therefore UhIlG" (h + G,,) is open; since this is the complement of G" in G it follows that G" is closed.

For topologies given by sequences of subgroups there is an alternative purely algebraic definition of the completion which is often convenient. Suppose (xv) is a Cauchy sequence in G. Then the image of Xv in GIG" is ultimately constant, equal say to ~". If we pass from n + 1 to n it is clear that ~,,+ 1 1-+ ~" under the projection

/ 8. + 1 / G G,,+l -~ G G".

Thus a Cauchy sequence (x,) in G defines a coherent sequence (~,,) in the sense that

8"+l~"+l =~" for all n.

Moreover it is clear that equivalent Cauchy sequences define the same sequence (~,,). Finally, given any coherent sequence (~,,), we can construct a Cauchy sequence (x,,) giving rise to it by taking x" to be any element in the coset ~" (so that x" + 1 - x" E G ,,). Thus G can equally well be defined as the set of coherent sequences (~,,) with the obvious group structure.

We have now arrived at a special case of inverse limits. More generally, consider any sequence of groups {A,,} and homomorphisms

We call this an inverse system, and the group of all coherent sequences (a,,) (i.e., a" E A" and 8"+la,,+ 1 = a,,) is called the inverse limit of the system. It is usually written lim A", the homomorphisms 8" being understood. With this

+--notation we have

G~ ~GIG".

The inverse limit definition of G has many advantages. Its main drawback is that it presupposes a fixed choice of the subgroups G". Now we can have. different sequences of G" defining the same topology and hence the same com­pletion. Of course we could define notions of "equivalent" inverse systems but the merit of the topological language is precisely that such notions are already built into it.

The exactness properties of completions are best studied by inverse limits. First let us observe that the inverse system {GIG,,} has the special property that 8" + 1 is always surjective. Any inverse system with this property we shall call a surjective system. Suppose now that {A,,}, {B,,}, {e,,} are three inverse systems and that we have commutative diagrams of exact sequences

0--+ A"+l --+ B"+l --+ C,,+l --+ 0 t t t

o --+ A" --+ B" --+ e" --+ O.

104 COMPLETIONS

We shall then say that we have an exact sequence of inverse systems. The diagram certainly induces homomorphisms

o --+ lim A" --+ lim B" --+ lim C" --+ 0 +- +- +-

but this sequence is not always exact. However, we have

Proposition 10.2. If 0 --+ {An} --+ {B,,} --+ {C,,} --+ 0 is an exact sequence of inverse systems then

o --+ lim A" --+ lim B" --+ lim C" +- +- +-

is always exact. If, moreover, {An} is a surjective system then

o --+ lim A" --+ lim B" --+ lim C" --+ 0 +- +- +-

is exact.

Proof Let A = [1:=1 A" and define d A: A --+ A by dA(an) = a" - 8,,+1(a,,+1)' Then Ker d A ~ lim An. Define B, C and dB, de similarly. The exact sequence

+-of inverse systems then defines a commutative diagram of exact sequences

O--+A--+B--+C--+O aAt aBt aCt

O--+A--+B--+C--+O

and hence by (2.10) an exact sequence

0--+ Ker d A --+ Ker dB --+ Ker de --+ Coker d A --+ Coker dB --+ Coker de --+ O.

To complete the proof we have only to prove that

{A,,} surjective => d A surjective,

but this is clear because to show d A surjective we have only to solve inductively the equations -~-

x" - 8,,+1(X,,+1) = a"

for x" E A", given an E A". •

Remark. The group Coker d A is usually denoted by liml An> since it is a derived +-

fUnctor in the sense of homological algebra.

Corollary 10.3. Let 0 --+ G' --+ G ~ G" --+ 0 be an exact sequence of groups. Let G have the topology defined by a sequence {G,,} of subgroups, and give G', GW the induced topologies, i.e. by the sequences {G~ !l G,,}, {pG,,}. Then

is exact.

FILTRATIONS lOS

P~oof Apply (10.2) to the exact sequences

G' G G" 0-+ G' n G" -+ G" -+ pG" -+ O. •

In particular we can apply (10.3) with G' = G", then G" discrete topology so that G" = G". Hence we deduce

Corollary 10.4. G" is a subgroup of G and

G/G" ;; G/G". •

Taking inverse limits in (l0.4) we deduce

Proposition 10.5. (i;; G. •

G/G" has the

If </>: G -+ G is an isomorphism we shall say that G is complete. Thus (10.5) asserts that the completion of G is complete. Note that our definition of com­plete includes Hausdorff (by (10.1»).

The most important class of examples of topological groups of the kind we are considering are given by taking G = A, G" = a", where a is an ideal in a ring A. The topology so defined on A is called the a-adic topology, or just the a-topology. Since the a" are ideals, it is not hard to check that with this topology A is a topological ring, i.e. that the ring operations are continuous. By (10.1) the topology is Hausdorff -¢> n a" = (0). The completion 1 of A is again a topo­logical ring; </>: A -+ 1 is a continuous ring homomorphism, whose kernel is na".

Likewise for an A-module M: take G = M, G" = a"M. This defines the a-topology on M, and the completion M of M is a topological A-module (i.e. 1 x M -+ M is continuous). Iff: M -+ N is any A-module homomorphism, then f(a"M) = a"f(M) £: a"N, and therefore f is continuous (with respect to the a-topologies on M and N) and so defines j: M -+ N. Examples. 1) A = k[x], where k is a field and x an indeterminate; a = (x). Then A = k[[x]], the ring of formal power series.

2) A = Z, a = (p), p prime. Then 1 is the ring of p-adic integers. Its elements are infinite series '2::'=0 a"p", 0 ~ an ~ p - 1. We have p" -+ 0 as n -+ 00.

FILTRATIONS

The a-topology of an A-module M was defined by taking the sub modules a"M as basic neighborhoods of 0, but there are other ways of defining the same topology. An (infinite) chain M = Mo ;2 Ml 2···;2 M" ;2 .•. , where the M" are submodules of M, is called afiltration of M, and denoted by (M,,). It is an a-filtration if aM" £: M" + 1 for all n, and a stable a-filtration if aM" = M ,,+1 for all sufficiently large n. Thus (a"M) is a stable a-filtration.

106 COMPLETIONS

Lemma 10.6. If(M,,), (M~) are stahle a-filtrations of M, then they have bounded difference: that is, there exists an integer no such that M"+,,o s M~ and M~+no s M"for all n ~ O. Hence all stable a-filtrations determine the same topology on M, namely the a-topology.

Proof Enough to take M~ = a"M. Since aM" s M,,+l for all n, we have a"M s M,,; also aM" = M,,+l for all n ~ no say, hence M"+,,o = a"M"o s a"M. _

GRADED RINGS AND MODULES

A graded ring is a ring A together with a family (A,,),,~o of subgroups of the additive group of A, such that A = 87:'-0 A" and AmA" S Am +" for all m, n ~ O. Thus Ao is a subring of A, and each A" is an Ao-module.

Example. A = k[x1, • •. , xr ), A" = set of all homogeneous polynomials of degree n. .

If A is a graded ring, a graded A-module is an A-module M together with a family (M")",,o of subgroups of Msuch that M = 87:'_oM"andAmM" S Mm+n for all m, n ~ O. Thus each M" is an Ao-module. An element x of M is homo­~eneous if x E M" for some n (n = degree of x). Any element y E M can be written uniquely as a finite sum L" y", where y" E M" for all n ~ 0, and all but a finite number of the y" are O. The non-zero components y" are called the homogeneous components of y.

If M, N are graded A-modules, a homomorphism of graded A-modules is an A-module homomorphismf: M --+ N such thatf(M,,) S N" for all n ~ O.

If A is a graded ring, let A+ = 87,,>0 A". A+ is an ideal of A.

Proposition 10.7. The following are equivalent, for a graded ring A:

i) A is a Noetherian ring;

ii) Ao is Noetherian and A is finitely generated as an Ao-algebra.

Proof i) => ii). Ao ~ A/A H hence is Noetherian. A+ is an ideal in A, hence is finitely generated, say by Xl> ... , x., which we may take to be homogeneous elements of A, of degrees kl> ... , k, say (all > 0). Let A' be the subring of A generated by Xl> ... , x. over Ao. We shall show that A" S A' for all n ~ 0, by induction on n. This is certainly true for n = O. Let n > 0 and let YEA". Since y E A+, y is a linear combination of the Xi> say Y = ~f.l alxl, where al E A,,-kl (conventionally Am = 0 if m < 0). Since each kl > 0, the inductive hypothesis shows that each al is a polynomial in the x's with coefficients in Ao. Hence the same is true of y, and therefore YEA'. Hence A" S A' and therefore A = A'.

ii) => i): by Hilbert's basis theorem (7.6). _

GRADED RINGS AND MODULES 107

Let A be a ring (not graded), a an ideal of A. Then we can form a graded ring A* = EB:'.o a". Similarly, if M is an A-module and M" is an a-filtration of M, then M* = EB" M" is a graded A*-module, since amM" s;; Mm+fI •

If A is Noetherian, a is finitely generated, say by Xl, ... , Xr; then A* = A[xl>' .. , xr] and is Noetherian by (7.6).

Lemma 10.8. Let A be a Noetherian ring, M a finitely-generated A-module, (M,,) an a-filtration of M. Then the following are equivalent:

i) M* is a finitely-generated A*-module;

ii) The filtration (M,,) is stable.

Proof Each M" is finitely generated, hence so is each Q" = EB~.o M r: this is a subgroup of M* but not (in general) an A*-submodule. However, it generates one, namely

M: = Mo EB· .. EB M" EB aM" EB a2 M" EB· .. EB arM" EB· ..

Since Q" is finitely generated as an A-module, M: is finitely generated as an A*-module. The M: form an ascending chain, whose union is M*. Since A* is Noetherian, M* is finitely generated as an A*-module -¢> the chain stops, i.e., M* = M:o for some no -¢> M"o+r = arM"o for all r ~ 0 -¢> the filtration is stable. _

Proposition 10.9. (Artin-Rees lemma). Let A be a Noetherian ring, a an ideal in A, M a finitely-generated A-module, (M,,) a stable a-filtration of M. If M' is a submodule of M, then (M' n M,,) is a stable a-filtration ofM'.

Proof We have a(M' n M,,) s;; aM' ('\ aM" s;; M' ('\ M"+l> hence (M' n M,,) is an a-filtration. Hence it defines a graded A*-module which is a submodule of M* and therefore finitely generated (since A* is Noetherian). Now use (10.8). •

Taking M" = a"M we obtain what is usually known as the Artin-Rees lemma:

Corollary 10.10. There exists an integer k such that

(a"M) n M' = a"-" (a"M) n M')

for all n ~ k. •

On the other hand, combining (10.9) with the elementary lemma (10.6) we obtain the really significant version:

5*

Theorem 10.11. Let A be a Noetherian ring, a an ideal, M afinitely-generated A-module and M' a submodule of M. Then the filtrations aflM' and (a"M) n M' have bounded difference. In particular the a-topology of M' coincides with the topology induced by the a-topology of M. •

lOS COMPLETIONS

Remark. In this chapter we shall apply the last part of (10.11) concerning topologies. However, in the next chapter the stronger result about bounded differences will be needed.

As a first application of (10.11) we combine it with (10.3) to get the im­portant exactness property of completion:

Proposition 10.12. Let

0--+ M' --+ M --+ M" --+ 0

be an exact sequence of finitely-generated modules ot'er a Noetherian ring A. Let a be an ideal of A, then the sequence of a-adic completions

0--+ M' --+ M --+ M" --+ 0

is exact. _

Since we have a natural homomorphism A --+ A we can regard A as an A-algebra and so for any A-module M we can form an A-module A ®A M. It is natural to ask how this compares with the A-module M. Now the A-module homomorphism M --+ M defines an A-module homomorphism

A 0 A M --+ A 0 A M --+ A ® A M = M.

In general, for arbitrary A and M, this is neither injective nor surjective, but we do have:

Proposition 10.13. For any ring A, if M isfinitely-generated, A ®~ M --+ M is surjective. If, moreover, A is Noetherian then A 0 A M --+ M is an iso­morphism.

Proof Using (10.3) or otherwise it is clear that a-adic completion commutes with finite direct sums. Hence if F ~ An we have A ®A F ~ F. Now assume M is finitely generated so that we have an exact sequence

o --+ N --+ F --+ M --+ O.

This gives rise to the commutative diagram

A ®A N --+ A 0 A F --+ A ®A M --+ 0 -p -J,8 -J,a

O--+N --+ F ~ M--+O

in which the top line is exact (by (2.18)). By (10.3) 8 is surjective. Since f1 is an isomorphism this implies that ex is surjective, proving the first part of the pro­position. Assume now that A is Noetherian, then N is also finitely generated so that y is surjective and, by (10.12), the bottom line is exact. A little diagram chasing now proves that ex is injecth;e and so an isomorphism. _

GRADED RINGS AND MODULES 109

Propositions (10.12) and (10.13) together assert that the functor M f-+

A ®A M is exact on the category of finitely-generated A-modules (when A is Noetherian). As shown in Chapter 2 this proves:

Proposition 10.14. If A is a Noetherian ring, a an ideal, A the a-adic com­pletion of A, then A is a flat A-algebra. -

Remark. For non-finitely-generated modules the functor M f-+ Ai is not exact: the good functor, which is exact, is M f-+ A ®A M and the two functors coincide on finitely-generated modules.

We proceed now to study the ring A in more detail. First some elementary propositions:

Proposition 10.15. If A is Noetherian, A its a-adic completion, then

i) a = Aa ~ A ®A a;

ii) (a"Y" = (a)";

iii) a"/a"+l ~ a"/a."+l;

iv) a is contained in the Jacobson radical of A. Proof Since A is Noetherian, a is finitely-generated. (10.13) implies that the map

A Q9A a~a,

whose image is ..la, is an isomorphism. This proves i). Now apply i) to a" and we deduce that

(a")~ = Aa" = (Aa)" = (a)"

Applying (10.4) we now deduce

A/an ~ A/a."

by (1.18) by i).

from which iii) follows by taking quotients. By ii) and (10.5) we see that A is complete for its a-topology. Hence for any x E a

(1 - x) - 1 = I + x + x 2 + ...

converges in A, so that 1 - x is a unit. By (1.9) this implies that a is contained in the Jacobson radical of A. _

Proposition 10.16. Let A be a Noetherian local ring, m its maximal ideal. Then the m-adic completion A of A is a local ring with maximal ideal m.

Proof By (10.15) iii) we have A/m ~ A/m, hence A/m is a field and so m is a maximal ideal. By (10.15) iv) it follows that m is the Jacobson radical of A and so is the unique maximal ideal. Thus A is a local ring. _

The important question of how much we lose on completion is answered by Krull's Theorem:

110 COMPLETIONS

Theorem 10.17. Let A be a Noetherian ring, a an ideal, M afinitely-generated A-module and M the a-completion of M. Then the kernel E = n:~1 a"M of M --+ M consists of those x E M annihilated by some element of 1 + a.

Proof Since E is the intersection of all neighborhoods of 0 E M, the topology induced on it is trivial, i.e., E is the only neighborhood of ° E E. By (10.11) the induced topology on E coincides with its a-topology. Since aE is a neighborhood in the a-topology it follows that aE = E. Since M is finitely-generated and A is Noetherian, E is also finitely-generated and so we can apply (2.5) and deduce from aE = E that (1 - a)E = 0 for some a E a. The converse is obvious: if (1 - a)x = 0, then

'" x = ax = a2x =. ··E n a"M = E. _ "-1

Remarks. 1) If S is the multiplicatively closed set 1 + a, then (10.17) asserts that

A --+ A and A --+ S"1A

have the same kernel. Moreover for any a E a (l - a) -1 = I + a + a 2 + ...

converges in A, so that every element of S becomes a unit in A. By the universal property of S-1A this means that there is a natural homomorphism S-1A --+ A and (10.17) implies that this is injective. Thus S-1A can be identified with a subring of A.

2) Krull's Theorem (10.17) may be false if A is not Noetherian. Let A be the ring of all Coo functions on the real line, and let a be the ideal of allf which van­ish at the origin (a is maximal since A/a ~ R). In fact a is generated by the identity function x, and n:", 1 a" is the set of all f E A, all of whose derivatives vanish at the origin. On the other handfis annihilated by some element 1 + a

(a E a) if and only if/vanishes identically in some neighborhood ofO. The well­known function e- 1IX2

, which is not identically zero near 0, but has vanishing derivatives at 0, then shows that the kernels of

A --+ A and A --+ S - 1 A (S = 1 + a)

do not coincide. Thus A is not Noetherian.

Krull's Theorem has many corollaries:

Corollary 10.18. Let A be a Noetherian domain, a -# (1) an ideal of A. Then n a" = O.

Proof I + a contains no zero-divisors. _

Corollary 10.19. Let A be a Noetherian ring, a an ideal of A contained in the Jacobson radical and let M be a finitely-generated A-module. Then the a-topology of M is Hausdorff, i.e.,n a"M = O.

THE ASSOCIATED GRADED RING 111

Proof By (1.9) every element of 1 + a is a unit. _

As a particularly important special case of (10.19) we have:

Corollary 10.20. Let A be a Noetherian local ring, m its maximal ideal, M a finitely-generated A-module. Then the m-topology of M is Hausdorff. In particular the m-topology of A is Hausdorff. _

We can restate (10.20) slightly differently if we recall that an m-primary ideal of A is just any ideal contained between m and some power m" (use (4.2) and (7.14»). Thus (10.20) implies that the intersection of all m-primary ideals of A is zero. If now A is any Noetherian ring, p a prime ideal, we can apply this version of (1 0.20) to the local ring Ap. Lifting back to A and using the one-to-one correspondence (4.8) between p-primary ideals of A and m-primary ideals of All (where m = -\JAp) we deduce:

Corollary 10.21. LetA be a Noetherian ring, -\J a prime ideal of A. Then the intersection of all p-primary ideals of A is the kernel of A --+ Ap. _

THE ASSOCIATED GRADED RING

Let A be a ring and a an ideal of A. Define

GO

G (A)( = Ga(A» = ffi a"/a" + 1 _ n=rO

(aO = A).

This is a graded ring, in which the multiplication is defined as follows: For each x" E a", let x" denote the image of x" in a"/an+l; define xmx" to be

XmX", i.e., the image of XmX" in am+n/am+"+l; check that xmx" does not depend on the particular representatives chosen.

Similarly, if M is an A-module and (M,,) is an a-filtration of M, define

co

G(M) = ffi M,,/Mn+l ".0

which is a graded G(A)-module in a natural way. Let G,,(M) denote M,,/Mn+l.

Proposition 10.22. Let A be a Noetherian ring, a an ideal of A. Then

i) Ga(A) is Noetherian;

ii) Ga(A) and Ga(A) are isomorphic as graded rings;

iii) if M is a finitely-generated A -module and (M ,,) is a stable a-jiltration of M. then G(M) is a finitely-generated graded Ga(A)-module.

Proof i) Since A is Noetherian, a is finitely generated, say by Xl> ••• , X,. Let XI be the image of XI in a/a2, then G(A) = (A/a)[xl> . .. , x.]. Since A/a is Noetherian, G(A) is Noetherian by the Hilbert basis theorem.

ii) a"/an+l ~ n"/n" + 1 by (10.15) iii).

112 COMPLETIONS

iii) There exists no such that M no + r = aT M no for all r ~ 0, hence G(M) is generated by EBn .. no Gn(M). Each Gn(M) = Mn/Mn+l is Noetherian and annihilated by a, hence is a finitely-generated A/a-module, hence EBn .. no Gn(M) is generated by a finite number of elements (as an A/a-module), hence G(M) is finitely generated as a G(A)-module. _

The last main result of this chapter is that the a-adic completion of a Noetherian ring is Noetherian. Before we can proceed to the proof we need a simple lemma connecting the completion of any filtered group and the associ­ated graded group.

Lemma 10.23. Let 4>: A -+ B be a homomorphism of filtered groups, i.e. 4>(An) £ Bn, and let G(4)): G(A) -+ G(B), ~: A -+ jj be the induced homo­morphisms of the associated graded and completed groups. Then

i) G(4)) injective => ~ injective;

ii) G(4)) surjective => ~ surjective.

Proof Consider the commutative diagram of exact sequences

0-+ An/An+l -+ A/An+l -+ A/An -+ 0 tGn(<;I» t an + 1 tan

0-+ Bn/Bn+l -+ B/Bn+l -+ B/Bn -+ O.

This gives the exact sequence

0-+ Ker Gn(4)) -+ Ker an+ 1 -+ Ker an -+ Coker GnC4» -+ Coker an+ 1

-+ Coker an -+ O.

From this we see, by induction on n, that Ker an = 0 (case i)) or Coker an = 0 (case ii)). Moreover in case ii) we also have Ker an + 1 -+ Ker an surjective. Taking the inverse limit of the homomorphisms an and applying (10.2) the lemma follows. _

We can now form a result which is a partial converse of (10.22) iii) and is the main step in showing that A is Noetherian.

Proposition 10.24. Let A be a ring, a an ideal of A, M an A-module, (Mn) an a-filtration of M. Suppose that A is complete in the a-topology and that Mis Hausdorff in its filtration topology (i.e. that nn Mn = 0). Suppose also that G(M) is a finitely-generated G(A)-module. Then M is a finitely-generated A-module.

Proof Pick a finite set of generators of G(M), and split them up into their homogeneous components, say ~I (1 ~ i ~ v) where ~l has degree say n(i), and is therefore the image of say Xl E Mn(j)' Let pi be the module A with the stable a-filtration given by F~ = ak+n(l) and put F = EBf=l Fl. Then mapping the generator 1 of each pi to Xl defines a homomorphism

4>; F -+ M

EXERCISES 113

of filtered groups, and G(4)): G(F) -+ G(M) is a homomorphism of G(A) modules. By construction it is surjective. Hence by (10.23) ii) 1> is surjective. Consider now the diagram

F-tM

at t 8

A ~ A

F-+M

Since F is free and A = A it follows that a is an isomorphism. Since Mis Hausdorff f3 is injective. The surjectivity of 1> thus implies the surjectivity of 4>, and this means that Xl' ... , X T generate M as an A-module. _

Corollary 10.25. With the hypotheses of (10.24), if G(M) is a Noetherian G(A)-module, then M is a Noetherian A-module.

Proof We have to show that every submodule M' of M is finitely generated (6.2). Let M~ = M' n Mn; then (M~) is an a-filtration of M', and the em­bedding M~ -+ Mn gives rise to an injective homomorphism M~/M~+l-+ Mn/Mn+ 1> hence to an embedding ofG(M') in G(M). Since G(M) is Noetherian, G(M') is finitely generated by (6.2); also M' is Hausdorff, since n M~ £

n Mn = 0; hence by (10.24) M' is finitely generated. _

We can now deduce the result we are after:

Theorem 10.26. If A is. a Noetherian ring, a an ideal of A, then the a­completion A of A is Noetherian.

Proof By (10.22) we know that

Ga(A) = Ga(.1)

is Noetherian. Now apply (10.25) to the complete ring A, taking M = A (filtered by an, and so Hausdorff). _

Corollary 10.27. If A is a Noetherian ring, the power series ring B = A[[xl , .•. , xn]] in n variables is Noetherian. In particular k[[x l , . .. , xn]] (k a field) is Noetherian.

Proof A [x 1> ••• , xn] is Noetherian by the Hilbert basis theorem, and B is its completion for the (Xl' ... , xn)-adic topology. _

EXERCISES

1. Let an: ZjpZ ~ Zjpnz be the injection of abelian groups given by an(t) = pn -1,

and let a: A ~ B be the direct sum of all the an (where A is a countable direct sum of copies of ZjpZ, and B is the direct sum of the Zjpnz). Show that the p-adic completion of A is just A but that the completion of A for the topology induced from the p-adic topology on B is the direct product of the ZjpZ. Deduce that p-adic completion is not a right-exact functor on the category of all Z­modules.

114 COMPLETIONS

2. In Exercise 1, let An = ex - l(pn B), and consider the exact sequence

0-+ An -+ A -+ AI An -+ O.

Show that lim is not right exact, and compute liml An. +- +-

3. Let A be a Noetherian ring, a an ideal and M a finitely-generated A-module. Using Krull's Theorem and Exercise 14 of Chapter 3, prove that

ao n an M = n Ker (M -+ Mm), nal m;2a

where m runs over all maximal ideals containing a. Deduce that

M = 0 ¢> Supp (M) (') yea) = 0 (in Spec (A».

[The reader should think of M as the "Taylor expansion" of M transversal to the subscheme V(a): the above result then shows that M is determined in a neighborhood of yea) by its Taylor expansion.]

4. Let A be a Noetherian ring, a an ideal in A, and A the a-adic completion. For any x E A, let x be the image of x in A. Show that

x not a zero-divisor in A =- x not a zero-divisor in A.

Does this imply that

A is an integral domain =- A is an integral domain?

[Apply the exactness of completion to the sequence 0 -+ A ::... A.]

S. Let A be a Noetherian ring and let a, 0 be ideals in A. If M is any A-module, let Ma, MO denote its a-adic and o-adic completions respectively. If M is finitely generated, prove that (Ma)o ~ Ma+o. [Take the a-adic completion of the exact sequence

0-+ omM -+ Mjbm M -+ 0

and apply (10.13). Then use the isomorphism

lim (lim M/(anM + omM» ~ lim M/(anM + onM) +m- +;;- +;;-

and the inclusions (a + 0)2" S an + On S (a + o)n.]

6. Let A be a Noetherian ring and a an ideal in A. Prove that a is contained in the Jacobson radical of A if and only if every maximal ideal of A is closed for the a-topology. (A Noetherian topological ring in which the topology is defined by an ideal contained in the Jacobson radical is called a Zariski ring. Examples are local rings and (by (10.15)(iv» a-adic completions.)

7. Let A be a Noetherian ring, a an ideal of A, and A the a-adic completion. Prove that A is faithfully flat over A (Chapter 3, Exercise 16) if and only if A is a Zariski ring (for the a-topology).

EXERCISES 115

[Since A is flat over A, it is enough to show that

M - M injective for all finitely generated M ¢> A is Zariski;

now use (10.19) and Exercise 6.]

8. Let A be the local ring of the origin in Cn (i.e., the ring of all rational functions fig E C(Zl' ... , zn) with g(O) f:. 0), let B be the ring of power series in Zh ••• , z" which converge in some neighborhood of the origin, and let C be the ring of formal power series in Zh ... , Zn, so that A c: B c: C. Show that B is a local ring and that its completion for the maximal ideal topology is C. Assuming that B is Noetherian, prove that B is A-flat. [Use Chapter 3, Exercise 17, and Exercise 7 above.]

9. Let A be a local ring, m its maximal ideal. Assume that A is m-adically complete. For any polynomialf(x) E A [x], let/ex) E (Ajm)[x] denote its reduction mod. m. Prove Hensel's lemma: if f(x) is monic of degree n and if there exist coprime monic polynomials g(x) , Ii(x) E (Ajm)[x] of degrees r, n - r with lex) =

g(x)li(x), then we can lift g(x), Ii(x) back to monic polynomials g(x), hex) E A [x] such that f(x) = g(x)h(x). [Assume inductively that we have constructed g/c(x), h/c(x) E A[x] such that gk(x)hk(x) - f(x) E mk A [x]. Then use the fact that since g(x) and Ii(x) are coprime we can find apex), bp(x), of degrees ~ Il - r, r respectively, such that x P = ap(x)gk(X) + bp(x)lik(x), where p is any integer such that 1 ~ p ~ n. Finally, use the completeness of A to show that the sequences gk(X), hk(x) converge to the required g(x), h(x).]

10. i) With the notation of Exercise 9, deduce from Hensel's lemma that if lex) has a simple root a: E Ajm, then f(x) has a simple root a E A such that a: = a mod m.

ii) Show that 2 is a square in the ring of 7-adic integers. iii) Let f(x, y) E k[x, y], where k is a field, and assume that f(O, y) has y = ao

as a simple root. Prove that there exists a formal power series y(x) = 2;'~o anx" such thatf(x, y(x) = o.

(This gives the "analytic branch" of the curve f = 0 through the point (0, ao).)

11. Show that the converse of (10.26) is false, even if we assume that A is local and that A is a finitely-generated A-module. [Take A to be the ring of germs of Coo functions of x at x = 0, and use Borel's Theorem that every power series occurs as the Taylor expansion of some C'" function.]

12. If A is Noetherian, then A[[Xlo ... , x,,]] is a faithfully flat A-algebra. [Express A -->- A [[Xl, ... , x n ]] as a composition of flat extensions, and use Exercise 5(v) of Chapter 1.]

11

Dimension Theory

One of the basic notions in algebraic geometry is that of the dimension of a variety. This is essentially a local notion, and, as we shall show in this chapter, there is a very satisfactory theory of dimension for general Noetherian local rings. The main theorem asserts the equivalence of three different definitions of dimension. Two of these definitions have a fairly obvious geometrical content, but the third involving the Hilbert function is less conceptual. It has, however, many technical advantages and the whole theory becomes more streamlined if one brings it in at an early stage.

After dealing with dimension we give a brief account of regular local rings, which correspond to the notion of non-singularity in algebraic geometry. We establish the equivalence of three definitions of regularity.

Finally we indicate how, in the case of algebraic varieties over a field, the local dimensions we have defined coincide with the transcendence degree of the function field.

HILBERT FUNCTIONS

Let A = EB:ao An be a Noetherian graded ring. By (10.7) Ao is a Noetherian ring, and A is generated (as an Ao-algebra) by say Xl, ... , X., which we may take to be homogeneous, of degrees kl' ... , k. (all > 0).

Let M be a finitely-generated graded A-module. Then M is generated by a finite number of homogeneous elements, say mj (1 ~ j ~ t); let rj = deg mj. Every element of M", the homogeneous component of M of degree n, is thus of the form 'i.dj(x)mj> where hex) E A is homogeneous of degree n - rj (and therefore zero if n < rj). It follows that Mn is finitely generated as an Ao­module, namely it is generated by all glx)mj where glx) is a monomial in the XI of total degree n - rj. ,

Let ,\ be an additive function (with values in Z) on the class of all finitely­generated Ao-modules (Chapter 2). The Poincare series of M (with respect to ,\) is the generating function of '\(Mn), i.e., it is the power series

ex>

P(M, t) = I '\(Mn)t" E Z[[t]]. n~O

"£

HILBERT FUNCTIONS 117

Theorem 11.1. (Hilbert, Serre). P(M, t) is a rational function in t of the form f(t )/0: ~ 1 (1 - t k l), where f(t) E Z[t].

Proof By induction on s, the number of generators of A over Ao. Start with s = 0; this means that An = 0 for all n > 0, so that A = Ao and M is a finitely­generated Ao module, hence M n = 0 for all large n. Thus P(M, t) is a polynomial in this case.

Now suppose s > 0 and the theorem true for s - I. Multiplication by x. is an A-module homomorphism of Mn into M n + k ., hence it gives an exact sequence, say

o ~ Kn ~ Mn ~ M n+k• ~ L n+k, ~ o. (1)

Let K = EBn Kn, L = EBn Ln; these are both finitely-generated A-modules (because K is a submodule and L a quotient module of M), and both are anni­hilated by x., hence they are Ao[xl' ... , x. _ d-modules. Applying.\ to (1) we have, by (2.11)

.\(1(,.) - '\(M,,) + '\(Mn+k.) - .\(L"+k.) = 0;

multiplying by t n + k• and summing with respect to n we get

(1 - tk')P(M, t) = peL, t) - tk'P(K, t) + get) (2)

where get) is a polynomial. Applying the inductive hypothesis the result now follows. •

The order of the pole of P(M, t) at t = 1 we shall denote by d(M). It provides a measure of the "size" of M (relative to A). In particular d(A) is defined. The case when all k j = 1 is specially simple:

Corollary 11.2. If each k j = 1, then for all sufficiently large n, A(M,,) is a polynomial in n (with rational coefficients) of degree* d - l.

Proof By (11.1) we have A(Mn) = coefficient of t n inf(t)·(1 - t)- •. Cancel­ing powers of (1 - t) we may assume s = d and f(l) #- o. Suppose f(t) = "N k· L..k ~ 0 akt ; smce

ex>

~(d+k-l) (1 - t)-a = ~ d-l tk

k~O

we have N

A(Mn) = L>k (d+~=~-l) foralln ~ N. k=O

and the sum on the right-hand side is a polynomial in n with leading term (L ak)na-1/(d - I)! #- O. •

Remarks. 1) For a polynomial f(x) to be such that f(n) is an integer for all integers n, it is not necessary for fto have integer coefficients: e.g., tx(x + n. • We adopt the convention here that the degree of the zero polynomial is -1: also

that the binomial coefficient (~1) = 0 for n ;:: 0, and = 1 for n = - 1.

118 DIMENSION THEORY

2) The polynomial in (11.2) is usually called the Hilbert function (or poly­nomial) of M (with respect to A).

Returning now to the sequence (1) let us replace x. by any element x E Ak which is not a zero-divisor in M (i.e., xm = 0 with m EM::::> m = 0). Then K = 0 and equation (2) shows that

deL) = d(M) - 1.

Thus we have proved

Proposition 11.3. If x E Ak is not a zero-divisor in M then d(M/xM) =

d(M) - 1. •

We shall use (1Ll) in the case where Ao is an Artin ring (in particular, a field) and A(M) is the length I(M) of a finitely-generated Ao-module M. By (6.9) I(M) is additive.

Example. Let A = AO[XI' ... , x.], where Ao is an Artin ring and the Xj are independent indeterminates. Then An is a free Ao-module generated by the

monomials X~l . .. ~. where L: m l = n; there are (s+n-I) of these, hence s-I

peA, t) = (1 - t)-o.

We shall now consider the Hilbert functions obtained from a local ring by passing to the associated graded rings as in Chapter 10.

Proposition 11.4. Let A be a Noetherian local ring, m its maximal ideal, q an m-primary ideal, M a .finitely-generated A-module, (Mn) a stable q­

filtration of M. Then

i) M/ M n is of finite length,for each n ~ 0;

ii) for all sufficiently large n this length is a polynomial g(n) of degree ~ s in n, where s is the least number of generators of q;

iii) the degree and leading coefficient ofg(n) depend only on M and q, not on the filtration chosen.

Proof i) Let G(A) = EBn qn/qn+t, G(M) = EBn Mn/Mn+ l . Go(A) = A/q is an Artin local ring, say by (8.5); G(A) is Noetherian, and G(M) is a finitely­generated graded G(A)-module (10.22). Each Gn(M) = Mn/Mn+l is a Noetherian A-module annihilated by q, hence a Noetherian A/q-module, and therefore of finite length (since A/q is Artin). Hence M/Mn is of finite length, and

n

In = I(M/Mn) = 2: I(M,_dM,). (1) '=1

ii) If Xl, ... , x. generate q, the images XI of the Xj in q/q2 generate G(A) as an A/q-algebra, and each XI has degree I. Hence by (11.2) we have I (Mn/Mn + I) = fen) say, where fen) is a polynomial in n of degree ~ s - 1 for all large n.

DIMENSION THEORY OF NOETHERIAN LOCAL RINGS 119

Since from (1) we have I"u - I" = fen), it follows that I" is a polynomial g(n) of degree ~ s, for all large n.

iii) Let (M,,) be another stable q-filtration of M, and let g(n) = I (M/M,,). By (10.6) the two filtrations have bounded difference, i.e., there exists an integer no such that M,,+no £ M n, Mn+no £ Mn for all n ~ 0; consequently we have g(n + no) ~ g(n), g(n + no) ~ g(n). Since g and g are polynomials for all large n, we have limn _", g(n)/g(n) = 1, and therefore g, g have the same degree and leading coefficient. _

The polynomial g(n) corresponding to the filtration (qn M) is denoted by x~(n):

(for all large n).

If M = A, we write Xq(n) for xf(n) and call it the characteristic polynomial of the m-primary ideal q. In this case (11.4) gives

Corollary 11.5. For all large n, the length I(Ajqn) is a polynomial Xq(n) of degree ~ s, where s is the least number of generators of q. _

The polynomials Xq(n) for different choices of the m-primary ideal q all have the same degree, as the next proposition shows:

Proposition 11.6. If A, m, q are as above

deg xq(n) = deg Xm(n).

Proof We have m ;2 q ;2 mT for some r by (7.16), hence mn ;2 q" ;2 mT" and therefore

Xm(n) ~ Xq(n) ~ Xm(rn) for all large n.

Now let n -? 00, remembering that the x's are polynomials in n. •

The common degree of the Xq(n) will be denoted by d(A): in view of (11.2) this means we are putting d(A) = d(Gm(A)) where d(Gm(A)) is the integer defined earlier as the pole at t = 1 of the Hilbert function of Gm(A).

DIMENSION THEORY OF NOETHERIAN LOCAL RINGS

Let A be a Noetherian local ring, m its maximal ideal. Let SeA) = least number of generators of an m-primary ideal of A.

Our ambition is to prove that SeA) = d(A) = dim A. We shall achieve this by proving SeA) ~ d(A) ~ dim A ~ SeA). (11.5) and (11.6) together provide the first link in this chain:

Proposition 11.7. SeA) ~ d(A). •

Next we shall prove the analogue for local rings of (11.3). Note that this proof useS the strong version of the Artin-Rees lemma (not just the topological part).

120 DIMENSION THEORY

Proposition 11.8. Let A, m, q be as before. Let M be a finitely-generated A-module, x E A a non-zero-divisor in M and M' = M/xM. Then

deg X~' ~ deg X~ - 1.

Proof. Let N = xM; then N ~ M as A-modules, by virtue of the assumption on x. Let N n = N n qn M. Then we have exact sequences

0-+ N/Nn -+ M/qnM -+ M'/qnM' -+ O.

Hence, if g(n) = I(N/Nn), we have

g(n) - x~(n) + x~'(n) = 0

for all large n. Now by Artin-Rees (10.9), (N,,) is a stable q-filtration of N. Since N ~ M, (11.4) iii then implies that g(n) and x~(n) have the same leading term; hence the result. _

Corollary 11.9. If A is a Noetherian local ring, x a non-zero-divisor in A, then d(A/(x)) ~ d(A) - 1.

Proof. Put M = A in (11.8). _

We can now prove the crucial result:

Proposition 11.10. d(A) ~ dim A.

Proof. By induction on d = d(A). If d = 0 then I(A/mn) is constant for all large n, hence mn = mn+1 for some n, hence mn = 0 by Nakayama's lemma (2.6). Thus A is an Artin ring and dim A = O.

Suppose d > 0 and let Po C PI C ... C Pr be any chain of prime ideals in A. Let x E PI, X ¢ Po; let A' = A/po, and let x' be the image of x in A'. Then x' =ft 0, and A' is an integral domain, hence by (11.9) we have

d(A'/(x')) ~ d(A') - 1.

Also, 'if m' i'-the maximal ideal of A', A'/m'n is a homomorphic image of A/mn, hence I(A/mn) ~ I(A'/m'n) and therefore d(A) ~ d(A'). Consequently

d(A'/(x')) ~ d(A) - 1 = d - I.

Hence, by the inductive hypothesis, the length of any chain of prime ideals in A' lex') is ~ d - 1. But the images of PI' ... , Pr in A' lex') form a chain of length r - 1, hence r - 1 ~ d - 1 and consequently r ~ d. Hence dimA ~ d. _

Corollary 11.11. If A is a Noetherian local ring, dim A is finite. _

If A is any ring, P a prime ideal in A, then the height of P is defined to be the supremum of chains of prime ideals Po C PI c ... C Pr = P which end at p: by (3.13), height P = dim All' Hence, from (11.11):

DIMENSION THEORY OF NOETHERIAN LOCAL RINGS 121

Corollary 11.12. In a Noetherian ring every prime ideal has finite height, and therefore the set of prime ideals in a Noetherian ring satisfies the descending chain condition. •

Remark. Likewise we may define the depth of p, by considering chains of prime ideals which start at p: clearly depth P = dim A/p. But the depth of a prime ideal, even in a Noetherian ring, may be infinite (unless the ring is local). See Exercise 4.

Proposition 11.13. Let A be a Noetherian local ring of dimension d. Then there exists an m-primary ideal in A generated by d elements Xl' ••• , X a, and therefore dim A ~ SeA).

Proof Construct Xl! ... , Xa inductively in such a way that every prime ideal containing (Xl! ••• , XI) has height ~ i, for each i: Suppose i > 0 and Xl, .•• , XI-l

constructed. Let Pi (1 ~ j ~ s) be the minimal prime ideals (if any) of (Xl' •.• , XI -1) which have height exactly i. - 1. Since i-I < d = dim A = height m, we have m =1= Pi (1 ~j ~ s), hence m =1= Ui=l Pi by (1.11). Choose XI E m, XI ¢ U Pj, and let q be any prime containing (Xl' ••. , XI)' Then q

contains some minimal prime \ideal P of (Xl' •.• , XI -1)' If P = Pi for some j, we have XI E q, XI ¢ p, hence q :::::> P and therefore height q ~ i; if P =1= Pi (1 ~j ~ s), then height P ~ i, hence height q ~ i. Thus every prime ideal containing (Xl' ••• , XI) has height ~ i.

Consider then (Xl' •.. , xa). If P is a prime ideal of this ideal, P has height ~ d, hence P = m (for P em=> height P < height m = d). Hence the ideal (Xl> ..• , xa) is m-primary. _

Theorem 11.14. (Dimension theorem.) For any Noetherian local ring A the following three integers are equal:

i) the maximum length of chains of prime ideals in A;

ii) the degree of the characteristic polynomial Xm(n) = I(A/mn);

iii) the least number of generators of an m-primary ideal of A.

Proof (11.7), (11.10), (11.13). •

Example. Let A be the polynomial ring k[xl , •.. , xn] localized at the maximal ideal m = (Xl>"" xn). Then Gm(A) is a polynomial ring in n indeterminates and so its Poincare series is (1 - t)-n. Hence, using the equivalence of (i) and (ii) in (11.14), We deduce that dim Am = n.

Corollary 11.15. dim A ~ dimk (m/m2). Proof If XI E m (1 ~ i ~ s) are such that their images in m/m2 form a basis of this vector space, then the XI generate m by (2.8); hence dimk (m/m2) = s ~ dim A by (11.13). •

if

Corollary 11.16. Let A be a Noetherian ring, Xl, ... , Xr E A. Then every minimal ideal P belonging to (Xl, •.. , xr) has height ~ r.

122 DIMENSION THEORY

Proof. In A~ the ideal (Xl> ... , xT) becomes +>"-primary, hence r ~ dim A~ = height +>. _

Corollary 11.17. (Krull's principal ideal theorem). Let A be a Noetherian ring and let X be an element of A which is neither a zero-divisor nor a unit. Then every minimal prime ideal +> of (x) has height 1.

Proof. By (11.16), height +> ~ 1. If height +> = 0, then +> is a prime ideal belonging to 0, hence every element of +> is a zero-divisor by (4.7): contra­diction, since x E +>. _

Corollary 11.18. Let A be a Noetherian local ring, x an element of m which is not a zero-divisor. Then dim A/(x) = dim A - 1.

Proof. Let d = dim A/(x). By (11.9) and (11.14) we have d ~ dim A-I. On the other hand, let XI (1 ~ i ~ d) be elements of m whose images in A/(x) generate an m/(x)-primary ideal. Then the ideal (x, Xl> ..• , Xli) in A is m­primary, hence d + 1 ~ dim A. _

Corollary 11.19. Let A be the m-adic completion of A. Then dim A =

dimA. Proof. A/m" ~ A/m" from (10.15), hence Xm(n) = xm(n). _

If Xl> ..• , Xli generate an m-primary ideal, and d = dim A, we call Xl' ••. , Xli

a system of parameters. They have a certain independence property described in the following proposition.

Proposition 11.20. Let Xl> ... , Xli be a system of parameters for A and let q = (Xl> ••• , Xej) be the m-primary ideal generated by them. Let f(tl' ... , tli)

be a homogeneous polynomial of degree s with coefficients in A, and assume that

f(x!> . .. , xa) E qS+1.

Then all the coefficients of f lie in m.

Proof. Consider the epimorphism of graded rings

a: (A/q)[tl , •.. , tli ] ~ Gq(A)

given by tl ~ XI> where tj are indeterminates and XI is Xj mod q. The hypothesis on f implies that J(tl, ... , t li) (the reduction of fmod q) is in the kernel of a. Assume if possible that some coefficient off is a unit, thenJ is not a zero-divisor (cf. Chapter 1, Exercise 3). Then we have

d(Gq(A)) ~ d(A/~)[tI" .. , tli]/(/)) becauseJE Ker (a) = d(A/q)[tlt ... , tli]) - 1 by (11.3) = d - 1 by the example following (11.3).

But d (Gq(A)) = d by the main theorem (11.14). This gives the required con­tradiction. _

REGULAR LOCAL RINGS 123

This proposition takes a simple form if A contains a field k mapping iso-morphically onto the residue field Aim:

Corollary 11.21. If k c A is a field mapping isomorphically onto AIm and if Xl' ... , Xa is a system of parameters, then Xl, ... , Xa are algebraically independent over k.

Proof Assume I(x!> ... , xa) = 0 where f is a polynomial with coefficients in k. Iff ~ 0 we can writef = fa + higher terms, where fa is homogeneous of degree s and fa ~ O. Apply (11.20) to fa and we deduce that fa has all its coefficients in m. Since J. has coefficients in k this implies J. == 0, a contradiction. Hence Xl, ... , Xa are algebraically independent over k. •

REGULAR LOCAL RINGS

In algebraic geometry there is an important distinction between singular and non-singular points (see Exercise 1). The local rings of non-singular points have as their generalization (to the non-geometric case) what are called regular local rings: these are rings satisfying any of the (equivalent) conditions i)-iii) of the next theorem.

Theorem 11.22. Let A be a Noetherian local ring of dimension d, m its maximal ideal, k = Aim. Then the following are equivalent:

i) Gm(A) ;:;; k[tl> ... , tal where the tj are independent indeterminates; ii) dimk (m/m2) = d;

iii) m can be generated by d elements.

Proof i) => ii) is clear. ii) => iii) by (2.8): see the proof of (11.15). iii) => i): let m = (Xl>"" xa), then by (11.20) the map cr.: k[Xl,"" xa] ~ Gm(A) is an isomorphism of graded rings. •

A regular local ring is necessarily an integral domain: this is a consequence of the following more general result.

Lemma 11.23. Let A be a ring, a an ideal- of A such that nn an = O. Suppose that Ga(A) is an integral domain. Then A is an integral domain.

Proof. Let X, y be non-zero elements of A. Then since n an = 0 there exist integers r, s ~ 0 such that X E aT, X ¢ a'+1, YEa', y ¢ aHl• Let X, y denote the images of x, y in GT(A), G.(A) respectively. Then x =1= 0, Y =1= 0, hence xy = x·y =1= 0, hence xy =1= O. •

Hence by (9.2) the regular local rings of dimension 1 are precisely the dis­crete valuation rings.

It can also be shown that if A is a local ring and Gm(A) is an integrally closed integral domain, then A is integrally closed. It follows that a regular local ring is integrally closed; but there are integrally closed local domains of dimension > I which are not regular.

124 DIMENSION THEORY

Proposition 11.24. Let A be a Noetherian local ring. Then A is regular if and only if A is regular.

Proof By (10.16), (10.26) and (11.19) we know that A is a Noetherian local ring of the same dimension as A and with tit as maximal ideal. Now use (10.22) which asserts that Gm(A) ~ Gfit(A) and the result follows. •

Remarks. 1) It follows from what we have said above that A is also an integral domain. Geometrically speaking this means that (locally)

non-singularity:::;> analytic irreducibility

or that, at a non-singular point, there is only one analytic "branch".

2) If A contains a field k mapping isomorphically onto Aim (the geometric case) then (11.22) implies that A is a formal power series ring over k in d in­determinates. Thus the completions of local rings of non-singular points on d-dimensiona1 varieties over k are all isomorphic.

Example. Let A = k[Xl' ... , xn] (k any field, Xj independent indeterminates); let m = (Xl>' •. , xn). Then Ant (the local ring of affine space kn at the origin) is a regular local ring: for Gm(A) is a polynomial ring in n variables.

TRANSCENDENTAL DIMENSION

We shall conclude this brief treatment of dimension theory by showing how the dimension of local rings connects up with the dimension of varieties defined classically in terms of the function field.

Assume for simplicity that k is an algebraically closed field and let V be an irreducible affine variety over k. Thus the coordinate ring A(V) is of the form

A(V) = k[Xl' ... , xn]/~

where ~ is a prime ideal. The field of fractions of the integral domain A(V) is called the field of rational functions on V and is denoted by keY). It is a finitely­generated extension of k and so has a finite transcendence degree over k-the maximum number of algebraically independent elements. This number is defined to be the dimension of V. Now recall that, by the Nullstellensatz, the points of V correspond bijective1y with the maximal ideals of A(V). If P isa point with maximal ideal m we shall call dim A(V)m the local dimension of V at P. We propose to prove

Theorem 11.25. For any irreducible variety V over k the local dimension of V at any point is equql to dim V.

Remark. We already know by (11.21) that dim V ~ dim Am for all m. The problem is to prove the opposite inequality, and for this purpose the main lemma is:

EXERCISES 125

Lemma 11.26. Let B £ A be integral domains with B integrally closed and A integral over B. Let m be a maximal ideal of A, and let n = m n B. Then n is maximal and dim Am = dim Bn.

Proof This is an easy consequence of the results of Chapter 5. First n is maximal by (5.8). Next if

(1)

is a strict chain of primes in A, its intersection with B is by (5.9) a strict chain of primes

(2)

This proves dim Bn ~ dim Am. Conversely given the strict chain (2) we can, by (5.16), lift this to a chain (1) (necessarily strict): thus dim Am ~ dim Bn. •

We can now proceed to:

Proof of (11.25). By the Normalization Lemma (Chapter 5, Exercise 16), we can find a polynomial ring B = k[x l , . .. , xa] contained in A(V) such that d = dim Vand A(V) is integral over B. Since B is integrally closed (remark following (5.12)) we can apply (11.26) and this reduces our task to proving (11.25) for the ring B, i.e. for affine space. But any point of affine space can be taken as the origin of coordinates and, as we have already seen, k[XI' ... , xa] localized at the maximal ideal (Xl' ... , xa) is a local ring of dimension d. •

Corollary 11.27. For every maximal ideal m of A(V) we have

dim A(V) = dim A(V)m.

Proof By definition we have dim A(V) = supm dim A(V)m. But by (11.25) all A(V)m have the same dimension. •

EXERCISES

1. Let fE k[xl>.'" xn] be an irreducible polynomial over an algebraically closed field k. A point P on the variety f(x) = 0 is non-singular -= not all the partial derivatives af/ax, vanish at P. Let A = k[xt. ... , xn]/(I), and let m be the maximal ideal of A corresponding to the point P. Prove that P is non-singular -= Am is a regular local ring. [By (11.18) we have dim Am = n - 1. Now

m/m2 ~ (Xl>"" xn)/(XL, ... , Xn)2 + (I)

and ~as dimension n - 1 if and only if f¢ (XL, ... , Xn)2.]

2. In (11.21) assume that A is complete. Prove that the homomorphism k[[tL, ... , tal] --+ A given by t, 1-+ Xj (l ~ i ~ d) is injective and that A is a finitely-generated module over k[[t1 , ••• , tal]. [Use (10.24).]

126 DIMENSION THEORY

3. Extend (11.25) to non-algebraically-closed fields. [If k is the algebraic closure of k, then k[Xlo ... , xn] is integral over k[Xh ... , xn].]

4. An example of a Noetherian domain of infinite dimension (Nagata). Let k be a field and let A = k[X1, X2, ... , Xn, ... ] be a polynomial ring over k in a countably infinite set of indeterminates. Let m1, m2, . " be an increasing sequence of positive integers such that m, + 1 - m, > m, - m'-l for all i > 1. Let +>, = (Xm,+l, ... , xm'+l) and let S be the complement in A of the union of the ideals +>,.

Each +>, is a prime ideal and therefore the set S is multiplicatively closed. The ring S -1 A is Noetherian by Chapter 7, Exercise 9. Each S -1+>, has height equal to m,+ 1 - mit hence dim S -lA = 00.

S. Reformulate (11.1) in terms of the Grothendieck group K(Ao) (Chapter 7, Exercise 26).

6. Let A be a ring (not necessarily Noetherian). Prove that

1 + dim A .;;; dim A[x] .;;; 1 + 2 dim A.

[Let!: A -- A[x] be the embedding and consider the fiber of/*: Spec (A[xD-+ Spec (A) over a prime ideal +> of A. This fiber can be identified with the spectrum of k ® AA [x] ;;; k[x], where k is the residue field at +> (Chapter 3, Exercise 21), and dim k[x] = 1. Now use Exercise 7(ii) of Chapter 4.]

7. Let A be a Noetherian ring. Then

dim A[x] = 1 + dim A,

and hence, by induction on n,

dim A[X1 •.. . , xn] = n + dim A.

[Let +> be a prime ideal of height m in A. Then there exist ab ... , am E +> such that +> is a minimal prime ideal belonging to the ideal a = (ab" ., am). By Exercise 7 of Chapter 4, +>[x] is a minimal prime ideal of a [x] and therefore height +>[x] .;;; m. On the other hand, a chain of prime ideals +>0 c: +>1 c: ... c: +>m = +> gives rise to a chain +>o[x] c: ... c: +>m[X] = +>[x], hence height +>[x] ~ m. Hence height +>[x] = height +>. Now use the argument of Exercise 6.]

Index

Additive function, 23 Adic topology, 105 Algebra, 30

homomorphism, 30 Annihilator ideal, 8, 19 Artinian, module, 74

ring, 76 Artin-Rees lemma, 107

Boolean ring, 11

Chain conditions, 74 Chain of submodules, 76 Cokemel,19 Complete, 105 Completion, 102 Composition series, 76 Constructible topology, 48 Contraction, 9

Dedekind domain, 95 Dimension, 90 Discrete valuation ring, 94 Domain, integral, 2

principal ideal, 5

Exact sequence, 22 Extension of an ideal, 9

Field, 3 residue, 4, 43

Filtration, 105 Finite, A-algebra, 30

type, 30 Finitely-generated, 30 Flat, 29

faithfully, 29 Fractions, ring of, 36

127

Generators of a module, 20 Grothendieck group, 88

Height, 120 Hensel's lemma, 115 Hilbert basis theorem, 81 Hilbert function, 118 Hilbert Nullstellensatz, 67, 69, 82, 85

Ideal(s), 2 coprime, 7 decomposable, 52 direct product of, 7 fractional, 96 generated by,S intersection of, 6 invertible, 96 maximal,3 primary, 50 prime, 3 principal, 11 product of, 6 quotient, 8 sum of, 6

Image, 2 Integral, A-algebra, 60

closure, 60 element, 59

Integrally closed domain, 60

Kernel, 2, 18

Length, 76 Limit, direct, 32

inverse, 103 Local ring, 4 Localization, 38

128 INDEX

Modular law, 6 Module(s), 17

direct product of, 20 direct sum of, 20 faithful, 20 finitely-generated, 20 fiat, 29 free, 21 graded, 106 homomorphism, 18 product of, 19 quotient, 18 sub, 18 sum of, 19

Multiplicatively closed set. 36

Nakayama's lemma, 21 Nilpotent, 2 Noetherian, module, 74

ring, 76 Normalization lemma, 69

Parameters, system of, 122 Poincare series, 116 Primary, decomposition, 51

ideal, 50 Prime ideal(s), 3

associated, 52 embedded, 52 isolated, 52

Radical, Jacobson, 5 nil, 5 of a sub module, 57

Regular local ring, 123

Ring, I absolutely fiat, 35 Boolean, 11 discrete valuation, 94 graded, 106 homomorphism, 2 local, 4 of fractions, 36 quotient, 2 semi-local, 4 sub, 2 valuation, 65

Saturated, 44 Scalars, extension of, 27

restriction of, 27 Spectrum, maximal, 14

prime, 12 Support, 46 Symbolic power, 56

Tensor product, of algebras, 30 of modules, 24

Torsion, element, 45 submodule, 45

Unit, 2

Valuation ring, 65 Varieties, affine algebraic, 15

Zariski, ring, 114 topology, 12

Zero-divisor, 2 Zorn's lemma, 3